You are on page 1of 5

Proceedings of the 26th Chinese Control Conference

July 26-31, 2007, Zhangjiajie, Hunan, China

Wireless Sensing and Control of Structural Vibration


from Earthquake*
Liu Lin1, Dyke Shirley J2 , Veto Rebecca2
1. School of Civil Engineering and Architecture, Beijing Jiaotong University, Beijing 100044, P.R.China
E-mail: lliu@bjtu.edu.cn
2. School of Engineering and Applied Science, Washington Uinversity, St. Louis, MO 63130, USA
E-mail: sdyke@seas.wustl.edu; rlv1@cec.wustl.edu
Abstract: This paper presents the feasibility of wireless sensing technology in seismic response control of structures. A
shaking table test of a 3-story steel structure using MICA2 motes with accelerometers is conducted at Washington University
Structural Control and Earthquake Engineering Lab. Magnetorheological (MR) dampers are adopted as actuators. Based on
wireless acceleration feedback of the structures and wired force feedback of MR dampers, a bang-bang clipped optimal
control algorithm is developed to command the MR damper. The experimental results indicate that the proposed control
scheme can effectively mitigate the response of the structure.
Key Words: Wireless Sensors, MR Dampers, Protective Systems, Structural Control, Shaking Table Test

decentralized manner, is promising. However,


application of wireless sensor networks in real-time
feedback control systems is still scarce.
The purpose of this paper is to demonstrate the
feasibility of wireless sensing technology in seismic
response control of buildings. A real-time shaking table
test of a 3-story steel structure using smart sensors and
smart dampers is conducted at the Washington
University Structural Control and Earthquake
Engineering Lab. In this test, MICA2 motes with
acceleration sensor are adopted as sensors; MR
dampers are used as semi-active actuators. The results
of a control experiment using the wired accelerometers
are compared with those using off-the-shelf wireless
accelerometers. A wireless sensor is installed on the
second and third floors of the tested structure to
measure the acceleration of the floors. The nominal
discrete controller design follows the matched
pole-zero method because of the low sampling rate of
the wireless sensors.

1 INTRODUCTION
Buildings are subjected to natural hazards, such as
earthquakes, winds, hurricanes, fire and crimes during
their long-term use. Control technologies have attracted
great interest from structural engineering community as
a means of disaster prevention and relief [1]. In order to
acquire real-time sensor data for control decisions,
cables are traditionally used to connect sensors with a
controller. For a typical low-rise building, the
installation of a commercial wire-based data acquisition
(DAQ) system can cost upwards of a few thousand
dollars per sensing channel. As the size of the control
system grows or the actuator density rises, additional
cabling may result in significant increases in
installation time and expense.
To reduce the high-cost and labor-intensive
installations, wireless communication technology can
serve as an alternative to provide real-time data links
among the nodes in a control system. Recently, wireless
sensors and sensor networks are emerging as a new
paradigm for the implementation of structural health
monitoring (SHM). In this field, Lynch gave a
summary review of the collective experience of
wireless sensors and sensor networks for monitoring
structural performance and health [2]; Ruiz-Sandoval et
al. reported their experiences using the MICA Mote
wireless sensing platform for SHM [3]; Based on
extensive experience using the MICA and MICA2
platforms, Spencer et al. identified critical hardware
and software issues that must be addressed before the
MICA Motes can be used for SHM [4-5]. A significant
benefit of wireless technology for civil engineering
applications is that they are much less expensive to
install and maintain because fewer cables are needed.
The possibility for dense sensor networks that can
process large amounts of data, most likely in a

2 EXPERIMENTAL SETUP
The experimental model is a 3-story steel frame
building (see Fig.1). The mass of each story is
simulated by steel plates, i.e., each of the first and
second floor consists of three plates: one 30.48 cm
50.8 cm0.95 cm (12 in20 in3/8 in) plate and two
30.48 cm38.1 cm1.27 cm (12 in15 in1/2 in) plates
with a total mass of 34.05 kg; the third floor consists of
one 30.48 cm50.8 cm0.95 cm (12 in 20 in3/8 in)
plate and one 30.48 cm38.1 cm1.27 cm (12 in
15 in1/2 in) plate with a total mass of 22.70 kg.
These masses are chosen so that the building would
yield natural frequencies within the working
frequency range of the wireless sensors. Each story is
supported by four 3.175 cm30.48 cm0.3175 cm
(1.25 in12 in 1/8 in) steel columns with fixed
connections. The stiffness of each floor is 29,700N/m.
Several experiments are previously conducted with
various configurations of this structure and wired

* This work is supported by ITR, National Science Foundation under


Grant CCr-0325529, USA.

194

control strategies [6-7]. The test model is installed on a


uniaxial seismic simulator, which consists of a
1.5m1.5m aluminum sliding table mounted on
high-precision, low-friction and linear bearings.

Fig.1

data acquisition. For the wireless tests, two MICA2


motes are mounted on the second and third floor to
measure and transmit the acceleration information for
the feedback to the controller. Each mote integrates a
microprocessor, a radio transceiver, and a sensor board
into one small platform. A MTS310CA sensor board
manufactured by Crossbow Technology, Inc.(see Fig.3),
which is used in this research, contains acceleration,
magnetic, light, temperature, and acoustic sensors as
well as a sounder. These MICA2 motes are
programmed to acquire acceleration data over a range
of -0.25 g to +0.25 g corresponding to 40 discrete
values of the 10 bit A/D converter on the board. Two
Moteiv TelosB motes are connected to a Notebook PC
(base station) and used to receive the data transmissions
from the MICA2 motes.

Experimental setup

Two shear-mode MR dampers are installed as actuators


between the ground and the first floor of the structure to
provide the control force (see Fig.2). The device
consists of a paddle which slips between two steel
plates that are mounted with a coil to generate the
magnetic field. The paddle has foam on each side which
is saturated with MR fluid. Current is supplied to each
MR damper by a Lord Corporation Rheonetic Box
Device Controller. As the magnetic field increases,
more force is required to slide the paddle between the
plates of the MR device. Two PCB 208B01 force
transducers, powered by PCB 484B06 Power Units, are
used to measure the force applied to the structure for
the feedback to the controller.

Fig.2

Fig.3

MTS310CA sensor board

A digital-to-analog converter is built to facilitate


connection of the sensor outputs from the base station
to the DSP-based real-time controller manufactured by
dSpace, Inc. for control action determination. The
output of the controller is the command voltage for the
MR damper.

3 IDENTIFICATION OF TEST STRUCTURE


The first step in the experiment is to obtain a
mathematical model of the structural system for control
design. Herein we implement an automated approach
that is developed specifically for control-oriented
structural modeling. The method is based on the
Eigensystem Realization Algorithm (ERA) and
integrates the results of this automated system
identification technique with an analytical model of the
structural system.
The system to be identified has two inputs (i.e., the
ground acceleration and control force input) and three
outputs (i.e., the absolute acceleration of each floor). To
obtain a realization of the structure which has the
frequencies observed in the experimental system, an
analytical model of the system is developed and
verified. The parameters used for this analytical model
are based on the physical dimensions of the members
and the materials. Fixed connections are assumed at
column-beam joints. The lumped parameter model
takes the form
MX + CX +KX = M 
xg +f
(1)

MR damper installation

Wired sensors are used for data acquisition and system


evaluation, as well as for the initial control test to
confirm the performance of the control system prior to
wireless control test. Four PCB 370A02 accelerometers
are used to measure the absolute accelerations of the
ground and each floor of the structure. A SIGLAB
20-42, 16 channel data acquisition system is used for
195

where X=[x1, x2, x3]T, x1, x2, x3 are the relative


displacements of the mass center of the 1st, 2nd, 3rd
floor, respectively; xg is the ground acceleration; f is

4 PERFORMANCE TEST

the vector of control forces; =[1,1,1]T; =[1,0,0]T ;


M=diag([m1, m2, m3]), m1= m2= 34.05 kg; m3= 22.7 kg;
and K=[k1+k2, -k2, 0; -k2, k2+k3, -k3; 0, -k3, k3], k1= k2=
k3=29.7 kN/m.
The ERA is applied to experimental data to determine
the eigenvalue problem of the test model. The identified
natural frequencies and damping ratios are fe=[2.28,
6.36, 8.99] (Hz), and he=[0.018, 0.010, 0.010],
respectively. The modal matrix can be obtained from
the eigenvectors of M -1K. The new modified stiffness
matrix is computed by
K r = M diag([2f e ]2 ) T
(2)
Note that this approach results in a model of the system
which maintains the mode shapes of the analytical
model, but has the frequencies of the experimental
system. In addition, the damping matrix is determined
to have the modal damping ratios which are identified
by ERA method as follows
C = M diag(2he [2f e ]T ) T
(3)
These updated stiffness matrix and damping matrix are
used to form the state space equations for this system as
z = AZ + Bf + Exg , y = Hz + Df
(4)
where z = [ X T , X T ]T is the state vector; and

To investigate the performance of MICA2 as the


wireless acceleration sensor, free vibration tests are
conducted. After some calibration test, the MICA2 and
a reference accelerometer (PCB 370A02 accelerometer)
are attached to the 3rd floor of the test structure.
Accelerations measured with the sampling rate of 50Hz
by the MICA2 are sent wirelessly to the base station,
which is attached directly to the notebook PC. It runs
on the TinyOS version 1.0 and has a re-try function for
sending information to the base station. Figure 5 shows
measured accelerations of both the reference
accelerometer and MICA2, and the acceleration
measured by the MICA2 agreed well with the result by
reference accelerometer. The sensitivity of the
accelerometer on MTS310CA Sensor Board is
sufficient for communication reliability. Only a small
fraction of data is lost during the test because of the
re-try function used for the wireless communication.
The accuracy of the acceleration measurements using
MICA2 is verified.

4.1

Wireless Sensors

y = [ 
x1a , 
x2 a , 
x3a ] is the vector of measured output

vector.
0
A=
1
M K r

I
0
0
, B = 1 , E =
1
M C
M

M 1 K r M 1C
M 1
H =
, D=

0
I

0
Fig.4 provides a representative transfer function
comparison of the identified model and the
experimentally obtained data. The transfer functions of
the model appear to well match the experimental data in
general. Small errors in the zeros of the transfer
functions may occur, but this is not expected to be
problematic for semi-active control systems.

Fig.5

4.2

Performance evaluation of MICA2 from free


vibration test

MR Dampers

In this study, MR dampers are used as semi-active


control actuators. Adequate modeling of actuators is
essential for adequate prediction of the behavior of the
controlled system. Dyke et al. proposed a Bouc-Wen
based model to accurately predict the behavior of a
shear-mode MR damper in their previous work [6-7]. The
governing equations of the dynamic model are
f = c0 xd + z
(5a)
z = xd z z

n 1

xd z + Axd
n

(5b)

= a + b u , c0 = c0 a + c0b u
(5c)
u = (u v)
(5d)
where f=the force generated by the damper; xd =the
stroke of the damper; c0=damping coefficient; z= the
evolutionary variable; v=the voltage applied to the
PWM circuit of the damper; by adjusting the
parameters , ,n and A, one can control the linearity in
the unloading and the smoothness of the transition from
the pre-yield to the post-yield region.
A constrained nonlinear optimization is used to obtain
these parameters. The optimization is performed using a

Fig.4 A typical transfer function of the test structure: from the


ground acceleration excitation to the acceleration of 3rd floor

196

sequential quadratic programming algorithm available


in MATLAB. The initial set of optimized parameters
that are determined to minimize the error between the
experimental results and the predicted response in a
variety of representative tests are: a=2730 N/m, b=2650
NV/m,c0a=3.2 Ns/m, c0b=2NsV/m, n=1, A=120,=3h
108m-3,=3h104m-1,=80 s-1.

5 SEMIACTIVE CONTROL ALGORITHM


In this study, a clipped-optimal control algorithm to
drive MR damper, which was first proposed by Dyke et
al.[8] and further verified by Jansen and Dyke [9], is
employed to determine the control action. The principle
of the clipped-optimal control algorithm for the MR
damper can be interpreted as follows: First, a nominal
active controller is assumed and designed as the
primary controller. Then, a secondary bang-bang-type
controller is designed to drive the MR damper through
the command voltage signal to emulate the control
force generated by the primary controller.
For the design of primary controller, Spencer et al.
successfully proposed a H2/LQG control design for
seismic protection [10]. The feedback measurements
included the acceleration of the 2nd and 3rd floor, as
well as measurements of forces provided by the MR
damper. In the design of H2/LQG controller, the ground
acceleration input, xg , is taken to be a stationary
white noise, an infinite horizon performance index is
chosen as

1
J = lim E { yrT Qyr + f T Rf }dt
(6)

where Q and R are weighing matrices for the vectors of


regulated responses yr = [ x1 , x2 , x3 , x1 , x2 , x3 ] and of
control force f, respectively. Additionally, R is a 1x1
identity matrix (R=1), and Q is the response weighting
matrix, Q = diag{[q1 I 33 , q2 I 33 ]} .
The nominal controller is represented as
z = ( A LC ) z + ( B LD) f m
(7)
f c = Kz
(8)
Where L is the gain matrix for the state estimator and K
is the gain matrix for the linear quadratic regulator.
In the experiment, the primary control design is a
discrete-time, feedback compensator of the form
xkc +1 = f1 ( xkc , ym , k , f m , k , k )
(9)
uk = f 2 ( xck , ym , k , f m , k , k )

Fig .6 Transfer function of the test structure for ideal active


control with various sampling rates and converting methods:
from the ground acceleration excitation to the acceleration of
3rd floor

The purpose of secondary controller is to determine the


command voltage to make the MR damper force
emulate the desired control force. In this paper, a
clipped optimal controller is used which follows the
bang-bang type control. The voltage applied to the MR
damper can be commanded as
vk = Vmax H ({uk f m , k 1} f m , k 1 )
(11)
where vk is the voltage applied at time t=kT; Vmax is
the maximum voltage of the damper, i.e., 4V in this
study; uk is the desired optimal control force at time
t=kT; f m , k 1 is the measured damper force at time
t=(k-1)T, respectively, H() is the Heaviside step
function.

6 EXPERIMENTAL RESULTS
As a basis for comparison, two cases are considered in
which the MR damper is employed in a passive mode.
In the first case, designated passive-off, the command
voltage to the MR damper is held at 0V. The second
passive case, in which the voltage to the MR damper is
held at the maximum voltage level (4 V), is denoted as
passive-on. Also, for semi-active control, both the
wireless-sensing- based control and traditional
wired-sensing-based control system are installed on the
test structure. So, totally four control strategies are
studied in the experiment, i.e., passive-off, passive-on,
wired-sensing-based controller, and wireless-sensingbased controller.
A random white noise (0~20 Hz) is used for the ground
excitation with RMS acceleration value of 139.6cm/s2
and time duration of 3 minutes. Also, parametric studies
are conducted to obtain the optimal values for
weighting factors of the primary controller. The optimal
weighting parameters in Eq.(6) are q1=107 and q2=1.
Tab.1 shows the experimental results of the four control
strategies. For the two passive control cases, it is noted
that the passive-on system results in larger maximum
accelerations than the passive-off system. Apparently,
choosing a passive device that produces the largest
damping forces may not be the most effective approach
to protective system design.
The result of the wired control test demonstrated that
the semi-active systems performed better than the two
passive control systems. In the subsequent test, wireless

(10)

where xkc , ym,k, fm,k and uk are the compensator state


vector, the measured acceleration responses, the
measured damper force, and the desired control force at
time t=kT.
In converting continuous-time controller to discretetime controller, the matched pole-zero method is
adopted because of the low sampling rate of the
wireless sensor, which is around 50 Hz. Figure 6 shows
the performance of matched pole-zero is better than
the Tustin method for low sampling rate for the
ideal active controller.
197

sensors are used for feedback while wired


accelerometers are adopted for data acquisition and
control system evaluation. As in the wired control
experiments, the controlled system is excited for three
minutes using ground acceleration with similar
frequency domain characteristics and RMS value.
The wireless control system is able to achieve
reasonable performance in comparison with the wired
system. However, the degradation in control
performance by using wireless sensors is observed
when compared with the wired control system. The
decline in performance results mainly from
communication latencies of MICA2 and data loss in
data transmission by MICA2.

be better that the motes read and transmit each sample


point immediately. Different communication strategies
could be used to improve this experiment in the near
future,.

ACKNOWLEDGEMENTS
This research is partially supported by NSF ITR Grant
CCR-0325529 and NSF Research Experience for
Undergraduates at Washington University Grant EEC0353718. The work was performed whilst the first
author worked as a visiting researcher at Washington
University. The authors would like to thank Pengcheng
Wang, a high school student supported by STARS
Program at UMSL funded by Pfizer Inc. and Solutia
Inc., for assisting the wireless control test, and
Professor Chengyang Lu and graduate student Fei Sun
at the Computer Science & Engineering Department of
Washington University for providing and making the
related hardware and software for the wireless control
test.

7DE506UHVSRQVHRIWKHWHVWFRQWUROV\VWHP
Control
strategy
No control
Passive-off
Passive-on
Wired
Controller
Wireless
Controller

Acceleration /(cmgs-2)
1st floor
2nd floor
3rd floor
56.1
63.3
71.3
39.5
45.0
54.4
61.2
59.8
64.5

Control
force/N

4.0
17.5

37.0

40.1

48.2

6.1

43.3

47.4

51.1

12.3

REFERENCES
[1] Soong T T,
Spencer Jr B F. Supplemental energy
dissipation: state-of-the-art and state-of-the-practice.
Engineering Structures, 2002,24 (3): 243-259.
[2] Lynch J P, Loh K A summary review of wireless sensors and
sensor networks for structural health monitoring. Shock and
Vibration Digest, 2005,38 (2): 91-128.
[3] Ruiz-Sandoval M E, Spencer Jr B F, Kurata N.
Development of a high sensitivity accelerometer for the
mica platform// Proceedings of the 4th International
Workshop on Structural Health Monitoring, Stanford, CA,
September 15-17, 2003:1027-1034.
[4] Spencer Jr B F, Ruiz-Sandoval M E, Kurata N. Smart
sensing technologies: opportunities and challenges.
Structural Control and Health Monitoring, 2004,11:
349-368.
[5] Kurata N, Spencer Jr B F, Ruiz-Sandoval M E. Risk
monitoring of buildings with wireless sensors networ.
Structural Control and Health Monitoring, 2005,12:
315-327.
[6] Yi F, Dyke S J, Carlson J D. Experimental verification of
multi-input seismic control strategies for smart dampers.
Journal of Engineering Mechanics, 2001,127 (11): 11521164.
[7] Yoshida O, Dyke S J, Giacosa L M, Truman K Z.
Experimental verification of torsional response control of
asymmetric buildings using MR Dampers. Earthquake
Engineering & Structural Dynamics, 2003,32 (13):
2085-2105.
[8] Dyke S J, Spencer Jr B F, Sain, M K, Carlson J D. Modeling
and control of magnetorheological dampers for seismic
response reduction. Smart Materials and Structures, 1996,5
(4): 565-575.
[9] Jansen L M, Dyke S J. Semiactive control strategies for MR
dampers: comparative study. Journal of Engineering
Mechanics, 2000,126 (8): 795-803.
[10] Spencer Jr B F, Suhardjo J, Sain M K. Frequency domain
optimal control strategies for aseismic protection. Journal of
Engineering Mechanics, 1994, 120 (1): 135-158.

7 CONCLUSION
The application of wireless sensors as presented in this
paper could prove to be a more economical and
attractive option than their wired counterparts for
structural control. To examine their performance for
protective systems against earthquake, the vibration
control test of a three-story structure with wired and
wireless sensors is conducted on the shaking table. MR
dampers are used as semi-active actuators. The clipped
optimal controllers are proposed and designed for MR
dampers. Once the best control scheme is identified
through the wired control tests, the experiment is
performed using the wireless ones with the same
control design parameters. The results of wireless
control are then compared to the results of the wired
control test, and the wireless-sensing based control can
effectively suppress the motion of the structure,
although control performance of the wireless control
system can still not come up to that of the wired control
system.
While this research has provided a greater
understanding of wireless sensors and their application
in control systems, many questions still remain
unsolved. In wireless transmission, some amount of
data is lost, but this project did not test the extent of the
data loss, nor did it relate the data loss to the
performance of the controller. Furthermore, the mote
used in this experiment first stores sensor data in the
flash memory and then send them to the base station
later, i.e., it transmits several seconds of recorded data
at a time; whereas, for the purpose of control, it would

198

You might also like