Beruflich Dokumente
Kultur Dokumente
Applications
M. Battaini
Department of Structural Mechanics
University of Pavia
I27100 Pavia - Italy
and
S.J. Dyke
Department of Civil Engineering
Washington University
St. Louis, MO 63130 - USA
Abstract. The objective of this work is to experimentally examine some practical issues relating
to the reliability and performance of active structural control systems. Using the newly constructed
shaking table in the Department of Structural Mechanics at the University of Pavia, experiments
were conducted employing an actively controlled, three story test structure. A DC-motor based
active mass driver system was used as the active control device. An H2 /LQG controller based on
acceleration feedback is applied to the system. The rst objective of this work to demonstrate the
ecacy of the acceleration feedback control strategy. Next, two experiments are conducted with
the goal of investigating fault tolerant control system design. In the rst case, the occurence of a
sensor failure is simulated experimentally, and the resulting controlled performance and robust-
ness characteristics are assessed. The second set of tests is to evaluate the ability of the controller
to reduce the structural responses when reduced information is available for control action deter-
mination. Both of these issues are important practical concerns in the design and implementation
of a structural control system.
Key words: Active Structural Control, Active Mass Driver, H2/LQG, Protective Systems,
Control-Structure Interaction, System Identication
1 Introduction
The application of vibration control systems to civil engineering structures has
been investigated in recent years to demonstrate the ecacy of these systems dur-
ing severe dynamic loadings such as earthquakes (for example, see Housner et al.,
1990, 1993, 1997). To date, much of the research in this eld has been directed
toward theoretical and experimental evaluations of various control mechanisms and
algorithms. Although a number of full-scale applications of active control are in
service in Japan, questions have been brought forth regarding the power require-
ments, reliability, capital and maintenance expenses, and stability of these systems
(Fujino et al., 1996). To move from laboratory tests to full scale implementation,
these issues must be addressed in a practical and systematic manner.
Before applying these systems to full-scale structures, one must anticipate di-
culties which may be encountered during implementation and determine the best
2 M. Battaini and S.J. Dyke
way to respond to these situations. One such diculty is the occurence of a compo-
nent failure. Because earthquakes are random events with potentially catastrophic
results, the solution to this question may be dierent for civil engineering appli-
cations than for various other applications. A system which remains eective even
when subjected to partial failures is known as a fault tolerant system.
In this paper, the issue of sensor failure, one specic type of component failure, is
addressed. The question of how to respond to a sensor failure must be considered
for the development of fault tolerant structural control systems. The solution to
this question eects the need for sensor redundancy, as well as the complexity
and cost of the control system. To oset the chances of a sensor failure leading to
an ineective or unstable controller, the system may include redundant sensors,
but the expense involved in duplicated sensors, and the associated algorithms and
equipment to switch between sensors, may be prohibitive. However, if the control
system is completely halted when a sensor failure occurs, no performance gains
can be achieved. Thus, it is desireable to continue operation of the control system
if an improvement in performance is expected.
For these investigations, it is assumed that there is a supervisory system to moni-
tor operation of the controller, and that this system has the ability to detect the
occurence of a sensor failure. However, once the failure is detected, how might the
supervisory control computer be programmed to respond. The possible alterna-
tives include 1) completely halting operation of the control system, 2) switching
to a duplicate sensor, 3) continuing normal operation (ignore the failure), and 4)
switching to an alternate control algorithm which does not require the measure-
ment corresponding to the failed sensor. The decision would likely be based on
the number and location of the sensor(s) that failed because this would determine
the stability characteristics and the performance of the resulting controlled system.
The purpose of this paper is to investigate the performance and robustness of
structural control systems with sensor failures through a set of experiments. The
experiments are performed at the structural control laboratory in the Department
of Structural Mechanics at the University of Pavia. In the study, a three story test
structure, equipped with a DC-motor based active mass driver (AMD) is employed.
An AMD system is chosen here because a large number of the existing full-scale
structural control applications use AMD systems. These investigations are restrict-
ed to the case in which the sensor failure is detected before an earthquake begins.
Controller changes and sensor failures that occur during an earthquake are expect-
ed to have a dierent eect on system performance, and require further analysis.
A nominal controller is designed and implemented on the experimental struc-
ture. An H2 /LQG optimal controller based on acceleration feedback is adopted
for these studies. This control algorithm is chosen based on previous success when
2 Experimental Setup
The experiment was performed in the newly constructed structural control labo-
ratory in the Department of Structural Mechanics at the University of Pavia. A
uniaxial shaking table was used to impart base accelerations to a single bay, three-
story, scaled test structure equipped with a DC-motor based active mass driver
to supply the control forces. In addition, the laboratory has a data acquisition
system and a real-time control implementation system. A diagram of the experi-
mental setup is shown in gure 1 and the facilities are described in the following
paragraphs.
xa2 - D/A
6
Digital Controller
xa1 -
6
? - A/D
xg
-
Shaking Table
Fig. 2. Photograph of Test Structure and Shaking Table (structure shown with braces).
Fig. 3. Photograph of AMD Consisting of Moving Cart, Gear Truck, and Potentiometer to
Measure the Cart Motion (designed and manufactured by Quanser Consulting, Inc.).
2
Gy1xg Gy1u 3
6 Gy2 xg Gy2 u 7
G (s) = 64 ...
6
..
.
7
7
5
(1)
Gy5xg Gy5u
where: xg is the base acceleration, u the command signal to the AMD and y =
[y1 ; y2 ; ::; y5 ] the output vector.
In the frequency domain approach employed herein, the rst step is to obtain
the experimental transfer functions of the system. The transfer functions corre-
sponding to the rst column of the transfer function matrix in (1) are obtained by
exciting the structure with the base, and setting the u command to zero. Here, a
band-limited white noise (0{12 Hz) signal was used to drive the shaking table. To
acquire the transfer functions in the second column, the same type of excitation
was used to drive the AMD, and the base was held xed. In this experiment, the
data were acquired with a sampling rate of 30 Hz, an anti-aliasing lter cut-o
frequency of 10 Hz, and suitable input gains.
Next, each of the transfer functions is modeled as a ratio of two polynomials
in the Laplace variable, s, and the poles and zeros of each transfer function are
obtained by performing a least-squares t of the model to the experimental trans-
fer functions. This procedure is implemented in a MATLAB code. In tting the
transfer functions, a common denominator should be assumed for each column of
the transfer function matrix.
Once each of the transfer functions have been modeled, each column of the transfer
function matrix is assembled into a state-space form, as
x_ 1 = A1x1 + B 1xg (2)
y = C 1 x1 + D1 xg
(4)
y = [C 1 C 2 ] x + D2 u + D1 xg
Equation 4 does not represent a minimal realization of the system because it has
repeated eigenvalues corresponding to the complex poles of the structure, and the
corresponding eigenvectors are not linearly independent. Thus, by condensing out
the redundant states, a model reduction is performed (Moore 1981), and a model
is found of the form
x_ r = Axr + Bu + E xg
y = C y xr + Dy u + F y xg +v (5)
z = C z xr + Dz u + F z xg
where the measured output vector is y = [xa1 ; xa2 ; xa3 ; xrA ; xaA ], and the regulated
output vector is z = [xa1 ; xa2 ; xa3 ; xrA ; xaA ; x1 ; x2 ; x3 ]. Notice that the equation for
the regulated output vector z may include any linear combination of the states of
the system. In this experiment, the regulated responses included the displacements
of the
oors of the structure relative to the ground. Although these measurements
are not available for measurement, they were rebuilt based on the accelerations,
and included in the regulated output vector. This reduced model is used as a basis
for the control design. Figure 5 provides a representative comparison between the
resulting identied model and the experimentally measured transfer functions.
Here, the rst three natural frequencies of the structural system are 1.23 Hz, 3.61
Hz and 5.62 Hz.
Note that an important aspect of the identication procedure adopted here is
that it automatically incorporates the actuator dynamics and the eects of control-
structure interaction. Neglecting these eects would reduce the eectiveness of the
control system, and could even result in instabilities in the controlled system. The
characteristic eects of this interaction, described in Dyke et al. (1995a), is clearly
shown in the transfer function from the AMD command signal, u, to the absolute
acceleration of the AMD, xaA , in gure 6. The closely spaced pairs of poles and
150
0
100
−20
50
Magnitude (dB)
Phase
−40 0
−50
−60
−100
−80
−150
−100 −200
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Frequency (Hz) Frequency (Hz)
Fig. 5. Comparison between the Experimental (dotted) and Analytical (solid) Transfer Functions
from u to xa3 : Magnitude (left) and Phase (right).
40 200
30
150
20
100
10
Magnitude (dB)
50
0
Phase
−10 0
−20
−50
−30
−100
−40
−150
−50
−60 −200
0 2 4 6 8 10 0 2 4 6 8 10
Frequency (Hz) Frequency (Hz)
Fig. 6. Comparison Between the Experimental (dotted) and Analytical (solid) Transfer Functions
from u to xaA: Magnitude (left) and Phase (right).
zeros result from the coupling of the AMD device and the structure, as discussed
in (Dyke et al., 1995ab, 1996).
Controller (K ) ?
d - z -
Structural System (P )
Input - -
u y
Output
u 6
Controller (K ) ?
where r and Q are weightings on the control action and regulated outputs, respec-
tively. The measurement noises are assumed to be identically distributed, statisti-
cally independent Gaussian white noise processes with the ratio SSxvgi xvig = 25. The
control signal u is calculated as:
u = kx^ (7)
where x^ is an estimate of the system state vector provided by a Kalman lter
(Spencer et al, 1997a,b), and k is the full state feedback gain vector.
Prior to implementation of the control designs, loop gain transfer function was
examined as an indication of the closed loop stability of the system (Dyke et
al., 1995b, 1996). A block diagram is shown in gure 8 which describes the loop
gain transfer function. Furthermore, comparing the experimental loop gain trans-
fer function to the theoretical transfer function also serves as a verication that
the controller, Multi-Q board, and control actuator are working properly.
Once the nal controller has been designed, it is implemented in a SIMULINK
code, converted to C code using the Real-Time Workshop, compiled, and down-
loaded to the Multi-Q board for control implementation.
Acceleration (g)
Acceleration (g)
0.02
0.02
0
0
−0.02 −0.02
−0.04 −0.04
−0.06 −0.06
−0.08 −0.08
−0.1 −0.1
0 10 20 30 40 50 60 0 5 10 15 20 25 30 35
Time (Sec) Time (Sec)
(a) (b)
0.1 0.1
0.08 0.08
0.06 0.06
0.04 0.04
Acceleration (g)
Acceleration (g)
0.02 0.02
0 0
−0.02 −0.02
−0.04 −0.04
−0.06 −0.06
−0.08 −0.08
−0.1 −0.1
0 10 20 30 40 50 60 0 10 20 30 40 50
Time (Sec) Time (Sec)
. (c) (d)
Fig. 9. Ground Acceleration for (a) El Centro Earthquake, (b) Tolmezzo Earthquake, (c) Bagnoli
Irpino Earthquake (d) Simulated Input.
the structure, and this controller was designed to achieve a signicant reduction
in all three structural modes, this controller does not reduce the peak responses
for this excitation as well as it does for the other inputs.
Although this controller has good performance, in this experiment the achievable
performance of this system appears to be limited by the stroke of the AMD rather
than any force capabilities of the control device. Based on observations during the
experiment, the authors believe that improved performance could be achieved with
a control device that has a larger stroke. Other authors have proposed the use of
nonlinear control algorithms which take into account the stroke limitations (see for
60 experimental
theoretical
40
Magnitude (dB)
20
−20
−40
−60
0 5 10 15
Frequency (Hz)
Fig. 10. Comparison between Measured and Theoretical Loop Gain Transfer Functions.
TABLE I
Uncontrolled (U) and Controlled (C) Peak Structural Accelerations.
example: Iemura and Igarashi, 1996; Nguyen et al., 1997; Agrawal and Yang, 1997).
The results provided for this controller are the basis for comparison for the studies
performed in the following sections. This controller is designated the \full con-
troller" in the sequel.
TABLE II
Uncontrolled (U) and Controlled (C) RMS Structural Accelerations.
0.08 0.15
uncontrolled uncontrolled
0.06 controlled controlled
0.1
0.04
0.05
Acceleration (g)
Acceleration (g)
0.02
0 0
−0.02 −0.05
−0.04
−0.1
−0.06
−0.15
−0.08 0 5 10 15 20 25 30 35
0 5 10 15 20 25 30 35 Time (sec)
Time (sec)
(a) (b)
Fig. 11. Uncontrolled and Controlled (a) xa2 Responses Due to Tolmezzo Earthquake, (b) xa3
Responses Due to Tolmezzo Earthquake.
20
Magnitude (dB)
10
−10
−20
−30
−40
0 2 4 6 8 10
Frequency (Hz)
Fig. 12. Comparison between Uncontrolled and Controlled Transfer Function Magnitude from
the Ground Acceleration to the Third Floor Absolute Acceleration.
occurs, and there is no extraneous noise or erroneous signal from the failed sensor.
In designing this controller for this experiment, no special eorts were made to
develop a controller which could take advantage of a sensor failure. This study
serves as an evaluation of the robustness of the controller against unforseen fail-
ures in the sensors.
The sensors are numbered as follows: sensor 1 = xa1 , sensor 2 = xa2 , sensor 3
= xa3 and sensor 4 = xaA . All combinations of sensor failures were simulated
numerically, and a portion these were then tested experimentally. In simulation
only two cases were found to be unstable. The two unstable cases included the
case in which sensors 2 and 3 had failed, and the case in which sensors 2, 3, and 4
had failed. Thus, because sensors 2 and 3 had failed in both of these cases, these
measurements seemed to be most signicant in the operation of this controller.
Based on the simulation results, only modest performance gains were achievable
when all but one sensor fails. Because these cases were not expected to provide
any new insights, they were not tested experimentally. The experimental results
for the tested cases have been designated Cases A{G, as dened in Table III. In the
table, an \x" below a given measurement indicates that the corresponding sensor
is operating. In each case, the structure was subjected to an El Centro earthquake
excitation. The results of the study are given in Table III.
From the results in Table III, when only a single sensor fails (cases B{E) the
performance is minimally eected. Also, note that the results of Case E are very
similar to those of Case A, indicating that the acceleration of the actuator (sensor
4) appears to be of little importance for the performance of the full controller
considered here. However, when the sensor that fails is the third
oor acceleration
TABLE III
Results of Sensor Failure Tests (x indicates sensor is operating).
TABLE IV
Experimental Results for Reduced Response Information Control
Designs.
8 Conclusions
In this study a series of experiments were conducted to investigate sensor failures
in a structural control system. The goal is to investigate how sensor failures eect
the performance of the system, and determine the best approach for responding to
the failure. These results are intended to provide information for the development
of fault tolerant structural control systems for civil engineering structures.
A three story test structure, controlled with a DC-motor based active mass driver,
was used in the experiments. The experiments were conducted in the Department
of Structural Mechanics at the University of Pavia. The ecacy of H2 /LQG con-
trol algorithms based on acceleration feedback was rst demonstrated on the test
structure, and the results of this successful control design were used as a basis for
comparison in the subsequent studies.
The rst study was conducted to investigate the performance and robustness of
the controlled system in the event of an unforseen failure in a sensor. This study
Acknowledgment
Establishment of the structural control laboratory in the Department of Structural
Mechanics of the University of Pavia has been made possible by the grant ASI -
ARS 96-189, coordinate by Professor Amalia Finzi of the Polytechnic of Milan.
Additionally, the authors would like to thank Professor F. Casciati of the University
of Pavia, for his support in completing this work, Professor B.F. Spencer, Jr. of
the University of Notre Dame for designing the test structure, Quanser Consulting,
Inc. of Ontario, Canada for designing and building the AMD, and the laboratory
technicians, F. Barzon, M. D' Adamo, and G. Sforzini, for constructing the test
structure.
Agrawal, A.K., and Yang, J.N., 1997. "Stability of Actively Controlled Civil Engineering Struc-
tures with Actuator Saturation," J. of Structural Engineering, ASCE, vol. 123, No. 4, pp.
505-512.
Battaini M., F. Casciati and L. Faravelli, 1997a. \System Reliability Approach to Safety Anal-
ysis of Controlled Structures," ,Proc. ESREL '97, International Conference on Safety and
Reliability, Lisbon, June 17{20.
Battaini M., Casciati F. and Faravelli L., 1997b. \Fuzzy Control of Structural Vibration. An
Active Mass System Driven by a Fuzzy Controller," Earthquake Engineering and Structural
Dynamics, Special Issue on the Benchmark Structural Control Comparison, (to appear).
Dyke S.J., Spencer Jr. B.F., Belknap A.E., Ferrel, K.J., Quast, P. and Sain, M.K, 1994. \Absolute
Acceleration Feedback Control Strategies for the Active Mass Driver," Proc. First World
Conference on Structural Control, Pasadena, California, August 3{5 1994, Vol.2, pp. TP1:51{
TP1:60.
Dyke S.J., Spencer Jr. B.F., Quast P. and Sain M.K, 1995a. \The Role of Control-Structure
Interaction in Protective System Design," J. Engrg. Mech., ASCE, Vol. 121, no. 2, pp. 322{
338.
Dyke S.J., Spencer Jr. B.F., Quast P., Sain M.K. Kaspari Jr. D.C. and Soong, T.T., 1995b.
\Acceleration Feedback Control of MDOF Structures," J. Engrg. Mech., ASCE, Vol. 122, No.
9, pp. 907{918.
Dyke S.J., Spencer Jr. B.F., Quast, P., Kaspari Jr. D.C. and Sain, M.K., 1996. \Implementation
of an Active Mass Driver Using Acceleration Feedback Control," Microcomputers in Civil
Engrg.: Special Issue on Active and Hybrid Structural Control , Vol.11, pp. 305{323.
Dyke S.J., 1996. Acceleration Feedback Control Strategies for Active and Semi-Active Control
Systems: Modeling, Algorithm Development, and Experimental Verication, Ph.D. Disserta-
tion ,University of Notre Dame, Notre Dame, Indiana, USA.
Fujino, Y., Soong, T.T., and Spencer Jr., B.F. \Structural Control: Basic Concepts and Appli-
cations," Proceedings of the ASCE Structures Congress, Chicago, IL. April 15{18, 1996, pp.
1277{1287.
Green M. and Limebeer D. J.N. 1995. Linear Robust Control, Prentice Hall, Englewood Clis,
New Jersey.
Housner, G.W. & S.F. Masri, Eds. 1990. Proc. of the U.S. National Workshop on Structural
Control Research, USC Publications No. M9013, University of Southern California.
Housner, G.W. & S.F. Masri, Eds. 1993. Proc. of the Int. Workshop on Structural Control, Univ.
of Southern California.
Housner, G.W., Bergman, L.A., Caughey, T.K. Chassiakos, A.G., Claus, R.O., Masri, S.F., Skel-
ton, R.E., Soong, T.T., Spencer Jr., B.F. and Yao, J.T.P., 1997. \Structural Control: Past,
Present, and Future," J. Engrg. Mech., ASCE, Special Issue, Vol. 123, No. 9, pp. 897{971.
Iemura, H. and Igarashi, 1996. \Nonlinear Active Control Experiment of a Real Size Frame
Structure," Proceedings of the Second International Workshop on Structural Control, Hong
Kong, December, pp. 241{252.
MATLAB, 1994. The Math Works, Inc. Natick, Massachusetts.
Moore B.B., 1981. \Principal Component Analysis in Linear Systems: Controllability, Observ-
ability, and Model Reduction," IEEE Trans. on Automatic Control, pp. 26{31.
nguy Nguyen, T., Jabbari, F. and de Miguel, D., 1997. "High Performance Control of Buildings
Under Seismic Excitation: Designs for Saturating Actuators," Proc. of the Eleventh VPI SU
Symposium on Structural Dynamics and Control, May.
Quast P., Sain M.K., Spencer Jr. B.F. and Dyke S.J., 1995. \Microcomputer Implementations
of Digital Control Strategies for Structural Response Reduction," Microcomputers in Civil
Engineering, vol. 10, pp. 13{25
SIMULINK,1994. The Math Works, Inc. Natick, Massachusetts.
Spencer Jr. B.F., Suhardjo J. and Sain M.K.,1994. \Frequency Domain Optimal Control for
Aseismic Protection," J. Engrg. Mech., ASCE,Vol. 120, No. 1, pp. 135{159.