Вы находитесь на странице: 1из 14

Journal of http://jvc.sagepub.

com/
Vibration and Control

Active multimodal vibration suppression of a flexible structure with piezoceramic sensor and actuator
by using loop shaping
V Sethi, MA Franchek and G Song
Journal of Vibration and Control 2011 17: 1994 originally published online 17 January 2011
DOI: 10.1177/1077546310393440

The online version of this article can be found at:


http://jvc.sagepub.com/content/17/13/1994

Published by:

http://www.sagepublications.com

Additional services and information for Journal of Vibration and Control can be found at:

Email Alerts: http://jvc.sagepub.com/cgi/alerts

Subscriptions: http://jvc.sagepub.com/subscriptions

Reprints: http://www.sagepub.com/journalsReprints.nav

Permissions: http://www.sagepub.com/journalsPermissions.nav

Citations: http://jvc.sagepub.com/content/17/13/1994.refs.html

>> Version of Record - Oct 27, 2011

OnlineFirst Version of Record - Jan 17, 2011

What is This?

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Article
Journal of Vibration and Control
17(13) 1994–2006

Active multimodal vibration suppression ! The Author(s) 2011


Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
of a flexible structure with piezoceramic DOI: 10.1177/1077546310393440
jvc.sagepub.com
sensor and actuator by using loop shaping

V Sethi, MA Franchek and G Song

Abstract
This paper represents active multimodal vibration control of a flexible beam structure with piezoceramic (PZT) actu-
ators and sensors using the loop shaping method. With surface-bonded PZT patch actuators and sensors, the flexible
beam has both sensing and actuating capacities. Due to its flat auto spectrum in the specified frequency range, the
Schroeder wave is used as an excitation signal for the non-parametric identification of the flexible beam structure.
The identified open loop model is then used for the closed loop design by using the loop shaping method based on the
extended sensitivity charts. A loop shaping compensator is designed to achieve multimodal vibration suppression.
Numerical results showed a reduction of 8 decibels for the first mode and 12–14 decibels for the second and third
modes. Experimental results closely match the simulation results. Furthermore, the results of loop shaping method are
compared with those of the methods of linear quadratic regulator and pole-placement control, which are designed based
on state space models via the parametric identification of the flexible beam. Comparisons show that the loop shaping
method is easier to design since a parametric identification is not required and requires less control effort while
maintaining the effectiveness in vibration suppression.

Keywords
Flexible structures, loop shaping control, multimodal control, piezoceramics, vibration suppression
Received: 21 January 2009; accepted: 16 November 2010

offer the advantage of being integrated with the struc-


1. Introduction
tures. In this investigation, piezoceramic or lead zirco-
Structural vibration suppression of flexible systems is nate titanate (PZT) actuators and sensors are used for
an increasing engineering concern. Because these flexi- active vibration suppression. Initial research of the
ble structures are often lightweight, underdamped, and piezoelectric materials for use in intelligent structures
increasingly being used in aerospace and other indus- was done by Crawley and de Luis (1987). Bailey and
tries, the need for active vibration control of these Hubbard (1985) investigated active vibration control of
multimodal structures becomes imperative. Classical a cantilever beam using piezoelectric polymer poly
feedback control of these structures is particularly vinylidene fluoride (PVDF). Hagood and Anderson
attractive because no adaptation time or reference (1992) addressed the issue of simultaneously sensing
signal is needed, and the effects of variations in the and actuating using piezoelectric materials.
system parameters can be reduced through sensitivity
reduction (Bode, 1945). This is the so-called robust per-
formance problem. Amongst the feedback control
Department of Mechanical Engineering, University of Houston, Houston,
strategies in time domain are pole placement and opti- TX, USA
mal control, and, in frequency domain, are various
Corresponding author:
compensation techniques. G Song, Department of Mechanical Engineering, University of Houston,
Vibration control using smart materials is gaining Houston, TX 77204–4006, USA
increasing attention, primarily because smart materials Email: gsong@uh.edu

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 1995

Control design of flexible structures often relies on smart materials. McFarlane and Glover (1992) devel-
accurate modeling of the system dynamics. Various oped a loop shaping design procedure using H1 syn-
approaches have been proposed to establish the theo- thesis for attitude control of a flexible space platform.
retical model of smart structures with embedded/sur- Loop shaping is used to shape the nominal plant sin-
face-bonded PZT actuators and sensors. A list of a gular values to give desired open loop properties at
few, but not limited to, includes analytical modeling frequencies of high and low loop gain. Active vibration
(Hagood and Anderson, 1992), the analysis of the suppression is a growing interest in the industrial appli-
modal response (Butler and Rao, 1996), and finite ele- cation due to the demand for higher accuracy and lower
ment model (Chopra, 1996; Lazarus and Crawley, acoustic noise levels.
1996). An alternative approach is to use the experimen- Launch vehicles often experience severe vibration
tal data to capture the input-output dynamics. This problems during the ascent phase due to intense acous-
approach is simplistic and often less labor intensive tic fields and mechanically transmitted disturbances. To
and time consuming than the analytical or finite ele- solve this vibration problem, Falangas et al. (1994)
ment model approach. Also, the experimental approach designed control methods for achieving active damping
takes care of the issue of the sensor and actuator on plate structures by means of piezoelectric actuators.
dynamics, driving amplifiers, signal conditioners, and In the proposed control methods, loop shaping is used
also some of the underlying assumptions in analyti- to specify closed-loop objectives in terms of require-
cal study. In the experimental study, two kinds of ments on the open-loop characteristics. Based on an
models, parametric and nonparametric, can be identi- H1 loop shaping synthesis procedure, Gouatarbes
fied using the data. Parametric identification like ARX, et al. (2007) designed a robust controller for a travelling
ARMAX, and Subspace method, requires good com- wave ultrasonic motor (TWUM) operated under
puting sources and effort; whereas on the other hand, extreme environmental conditions such as strong vibra-
the nonparametric identification can capture the system tions. To achieve the smooth landing and minimal
dynamics with much less effort. However, with the non- residual vibration as the recording head approaches
parametric identification, simulations and controller the target track, Ding et al. (2004)developed a settling
designs, like linear quadratic regulator (LQR) control, controller of a recording head for hard disk drives using
that directly require the system parameters, cannot be a PZT transducer. The fine actuator path is designed by
carried out. In such situations, loop shaping becomes loop shaping for superior performance of the overall
an effective tool for the control designer without using a system. Tuning approaches are less effective in vibra-
parametric model of the system to be the controller tion suppression when several dominant modes are
(Albert, 1999, Albert et al., 2000). involved. Den Hamer et al. (2005) developed a com-
Loop shaping technique is getting very popular bined loop shaping and root-locus tuning approach to
among researchers in the past decade. Kaitwanidvilai reduce several dominant modes with low-order control-
and Parnichkun (2004) used loop shaping technique to lers. Ahuja et al. (2008) utilized loop shaping technique
control the position of pneumatic servo system. Hsien along with a PZT actuator for active control of noise
et al. (1997) applied H1 loop shaping design procedure across an aluminum panel. Experimental results dem-
in servo controller design for a high-performance track- onstrated the capability of PZT actuators along with
ing control system with two flexibly-linked masses. the loop shaping technique in increasing the noise
The designed servo controller is implemented by a transmission loss across an aluminum panel corre-
digital signal processor (DSP) while the robustness of sponding to the modal frequencies for broadband exci-
the tracking performance subjective to internal uncer- tations of the panel. Kim et al. (1999) studied active
tainties and external disturbances is verified. Folcher structural acoustic control of a flexible plate using pie-
(2002) successfully implemented a underwater motion zoelectric actuators. In the proposed approach, an H1
controller using loop shaping technique which can deal controller is designed by using the loop shaping design
with the unknown disturbances, actuator saturation procedure to achieve robust acoustic control of the pro-
and plant uncertainties. Hamer et al. (2005) controlled posed system subjected to parameter variations and
the transient behavior of a mass-spring system using external disturbances.
both pole placement and loop shaping technique by Ample research is available for the implementa-
placing the poles of the system at a desired location tion of first dominant mode vibration control of struc-
based on an open loop compensator. Later, Park tures. These include works on pole placement control
et al. (2008) built an active vibration isolation system (Manning et al., 2000), LQR control (Blanguernon
(AVIS) using voice coil actuator using loop shaping et al., 1999), robust control (Fujimoto, 1995), PPF con-
control to improve the control action. Sethi et al. trol (Gu and Song, 2005), model reference control
(2008) implemented loop shaping control on a model- (Gu and Song, 2007) and sliding mode control (Song
frame structure using a piezo actuator and sensor as and Gu, 2007). As the aerospace and civil industries

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


1996 Journal of Vibration and Control 17(13)

embrace lighter and more flexible structures, the need Here,  denotes both parametric and unstructured
of simultaneous multimodal control increases because uncertainty due to modeling errors. The control
it is insufficient to just control the first dominant mode input, u(t), is the voltage output from the PZT actua-
of the structure. Some related multimodal control tors that are surface bonded to the flexible beam.
research includes H1 control by Kar et al. (2000), The system dynamics in transfer function form is writ-
semi-active energy rate control by Corr and Clark ten as
(2003), adaptive positive position feedback control by
Rew et al. (2002), and LQR control by Prakah-Asante yðt, Þ ¼ Gð p, ÞuðtÞ þ Hð p, Þd ðtÞ ð1Þ
and Craig (1994) and Sethi and Song (2004).
Additional papers discussing multimodal control are where G(p,) represents the transfer function from u to
listed in references (Bayon de Noyer and Hanagud, y, i.e., the dynamics between the control input and PZT
1998; Han et al., 1999; Kang et al., 1999; Li and sensor output, and H(p,) represents the transfer func-
Ulsoy, 1999; Oueini and Nayfeh, 1997). tion from d to y, i.e. the, dynamics between disturbance
This paper presents the study on multimodal vibra- and piezoceramic sensor output. Here all the measured
tion control of a flexible beam using the frequency signals and input signals are in volts, therefore the
domain based loop shaping method. The flexible transfer functions given in Equation (1) include the
beam is equipped with piezoceramic patch sensors dynamics of the sensors, actuators, and the physical
and actuators. In this paper, the approach to controller elements of the system.
design using the sensitivity charts is first introduced. In order to achieve the objective of vibration suppres-
Then, the multimodal vibration controller is designed sion of the first three modes, the following tasks need to
using sensitivity charts, which were obtained based on be performed. First, a robust controller design process
the non-parametric system identification of the beam, will be created where the controller is realized using
so as to reduce the sensitivity at the resonant frequen- common controller design tools. Second, a nonpara-
cies. The designed loop shaping controller is lastly metric system identification that provides the beam
implemented on the flexible beam for active vibration transfer function G(s,) needs to be performed. Third,
control. The experimental results are compared with using the proposed control design strategy, implementa-
those using some time domain based approaches such tion of the controller on the plant will be executed for
as the LQR control to demonstrate the effectiveness of verification purposes.
the proposed loop shaping controller.
3. Loop shaping feedback controller
2. Problem statement design process
The goal of this research is to reduce the vibration of As we intend to perform loop shaping for control
the first three bending modes of a flexible beam with design, it is desirable to use frequency domain repre-
PZT sensors and actuators. Section 4.1 will provide sentation for the transfer function and controller.
details about the flexible beam setup. The Laplace transformation of Equation (1) is as
A general block diagram representing such an open follows,
loop system is shown in Figure 1. The disturbance term
is represented by d(t), control input by u(t), the output, Yðs, Þ ¼ Gðs, ÞUðsÞ þ Hðs, ÞDðsÞ ð2Þ
d ðÞ
measured by a PZT sensor, by y(t,) and p ¼ .
dt where a denotes the parametric and unstructured
uncertainty in the model. Model uncertainties may be
due to the modeling errors, aging of the physical sys-
tems, machining tolerances, and changes in operating
Disturbance (V) d (t ) environments. A frequency domain based unity feed-
back controller can be designed such that UðsÞ ¼
Gc ðsÞ½Yðs, Þ  RðsÞ, where Gc ðsÞ is the controller in
H ( p, α )
the Laplace domain, and R(s) is the command reference
signal. Thus, with the control law, the block diagram is
Control Sensor represented by Figure 2 and the closed loop transfer
Input (V) Output (V) function is derived as
G ( p,α )
u (t ) y (t , α ) Gðs, ÞGc ðsÞ Hðs, Þ
Yðs, Þ ¼ RðsÞ þ DðsÞ:
1 þ Gðs, ÞGc ðsÞ 1 þ Gðs, ÞGc ðsÞ
Figure 1. Open loop block diagram of the system. ð3Þ

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 1997

which is also equivalent to


Disturbance (V) D(s)
   
Sð j!, Þ  ð!Þ8! 2
, 8: ð9Þ
G(s, a)

Control From Equation (9), it is inferred that only the


Input (V)
R(s) dynamics of the plant in terms of Gð j!, Þ needs to
Gc(s) G(s, a)
U(s, a) Y(s, a) be known, and the controller to be designed is Gc ð j!Þ.
Thus, from the above discussion, the control design
objective emerges as a sensitivity reduction problem.
These specifications are now given in terms of contin-
uous reduction at all frequencies or at desired discrete
Figure 2. Closed loop compensated system. frequencies.

The Equation (3) can be further simplified as


3.2. Sensitivity
Yðs, Þ ¼ Tðs, ÞRðsÞ þ Sðs, ÞHðs, ÞDðsÞ ð4Þ Bode first introduced sensitivity as it gives the relative
sensitivity of the closed loop transfer function
where Sðs, Þ and Tðs, Þ are the closed loop sensitiv- Tðs, Þ ¼ Gðs, ÞGc ðsÞ=ð1 þ Gðs, ÞGc ðsÞÞ to the plant
ity and complementary sensitivity transfer functions model error. For a single input single output system,
given by at a given frequency !,

1 Gðs, ÞGc ðsÞ T=T


Sðs, Þ ¼ and Tðs, Þ ¼ : lim ¼S ð10Þ
1 þ Gðs, ÞGc ðsÞ 1 þ Gðs, ÞGc ðsÞ G!0 G=G

In this research, due to the regulatory nature of the Sensitivity, also known as the closed loop sensitivity
control system, the reference input R(s) is assumed to transfer function, is denoted as Sðs, Þ ¼ 1=½1 þ Lðs, Þ,
be zero. Thus, the closed loop system is represented as where Lðs, Þ ¼ Gðs, ÞGc ðsÞ represents the open loop
Yðs, Þ ¼ Sðs, ÞHðs, ÞDðsÞ: ð5Þ transfer function.
It is well known that all electromechanical systems
are strictly proper (i.e. there are more poles than zeros)
3.1. Performance specifications because of the presence of inertial elements. Also, the
The performance specifications are typically given controllers designed should be strictly proper since rate
as the absolute upper bound on the allowable magni- saturation in actuators and sensor noise amplifications
fication of the disturbance D(s) or the attenuation are unavoidable issues when controlling physical sys-
that is relative to the open loop performance. The tems. Therefore, the open loop systems have at least
allowable magnification can be represented as ð!Þ. two more poles than zeros. Thus, the relative degree
Mathematically, this is stated as of all compensated open loop electromechanical sys-
  tems is lower bounded by 2.
Sð j!, ÞHð j!, Þ  ð!Þ8 2
, 8 ð6Þ
Bode quantified the impact and trade off of using
or equivalently, feedback control to reducing systems sensitivity. For
  a rational open loop transfer function, Lðs, Þ, with a
   
Sð j!, Þ   ð!Þ  8! 2
, 8 ð7Þ relative degree of two or more and n right half plane
Hð j!, Þ
poles at locations pi, then for closed loop stability the
where
denotes the frequency range of interest. sensitivity function must satisfy
Equation (7) leads us to determine and have a priori
Z X
n
knowledge of Gð j!, Þ and Hð j!, Þ. 1  
An alternative performance specification could lnSð j!, Þd! ¼  Reð pi Þ ð11Þ
 also 0 i¼1
be given as a desired reduction; say where ð!Þ is in
decibels, which is relative to open loop performance. where Re(pi) denotes the real part of pi.
This is written as The integral in Equation (11) is called the Bode sen-
sitivity integral and implies that sensitivity reduction
      through feedback control over one frequency range
Sð j!, ÞHð j!, Þ  Hð j!, Þ  ð!Þ8! 2
, 8
leads to sensitivity amplification over another fre-
ð8Þ quency range.

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


1998 Journal of Vibration and Control 17(13)

For stable plants with a relative degree of two, the for the frequencies that lie under the 0 dB contour, a
Bode sensitivity integral is disturbance signal gets amplified for a closed loop
Z system in comparison with the open loop system. For
1   the open loop frequency response for the frequencies
lnSð j!, Þd! ¼ 0, ð12Þ
0 that lie outside the 0 dB contour, a disturbance signal
gets attenuated for a closed loop system in comparison
implying that the area of sensitivity reduction must be with the open loop system. In the controller design pro-
equal to the area of sensitivity amplification. cess, by displaying the open loop system on the sensi-
Thus, a design trade off emerges for the sensitivity tivity charts, the designer can observe the effect of the
reduction problem. This is also the precise reason why controller design on the closed loop sensitivity of the
designers should minimize the amount of feedback con- system, and thus, can achieve the desired sensitivity
trol specified. For unstable plants, the unstable poles profile.
contribute to the increase
P in peak of sensitivity by an The Nyquist stability point is at 180 and at a 0 dB
amount equal to  ni¼1 Reð pi Þ. This is plausible as the location on the open loop phase-gain plot. Since the
system needs some feedback control to stabilize the phase angle is periodic, the extended sensitivity chart
unstable plant. can be drawn by repeating the constant sensitivity mag-
nitude curves every 360 . Using the extended sensitivity
chart, shown in Figure 4, a multimodal open loop
3.3. Sensitivity charts system can be analyzed. This is precisely what is
The Nichols chart relates the open loop frequency being used in this investigation for multimodal control
response functions to the closed loop complementary of the structure.
sensitivity function. Similarly, the sensitivity chart In this investigation, the open loop frequency
relates the open loop frequency response functions to response of the flexible beam will be shaped on the
the closed loop sensitivity function. For a unity nega- extended sensitivity chart using the classical frequency
tive feedback, the sensitivity is given by domain based design tools. The controller zeros and
poles are identified to ensure that the desired frequency
1 response is shaped and is in the proper regions so as to
Sð j!, Þ ¼ , Lð j!, Þ ¼ Gð j!, ÞGc ð j!Þ:
1 þ Lð j!, Þ achieve both performance and stability requirements.
ð13Þ
3.4. Frequency response function
Figure 3 shows the constant radius sensitivity curves
on the open loop gain and phase of Lð j!Þ plane. The
model development
0 dB sensitivity curve is shown as an inverted ‘‘U’’ con- The sensitivity chart is used to relate the performance
tour. As said earlier, there is a design trade off for the specifications to the open loop transfer function. To
sensitivity reduction problem. This is better explained further the objective of this research, it is necessary to
from Figure 4. For the open loop frequency response find the transfer function of the system in the frequency

Figure 3. The sensitivity chart. Figure 4. The extended sensitivity chart.

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 1999

domain. To accomplish this, one can use the analytical listed in Table 1. Figure 6 shows the experimental setup
modeling approach or the experimental approach. of the beam in a cantilevered configuration. The beam
However, to eliminate the effort associated in analytical has very low damping characteristics and exhibits sig-
modeling, this research uses a measured frequency nificant vibrations when excited.
response function (FRF) of the open loop system. For the vibration suppression problem, it is neces-
For the measurement of frequency response of the sary to place the actuator and sensors at such a location
open loop system, Gð j!Þ, Fourier transform is per- where the modes that are to be suppressed are both
formed on the input, u(t), and output signal, y(t). observable and controllable. This study utilizes PZT
From the block diagram of Figure 5, sensors, which measure the strain. Li and Ulsoy
(1999) have plotted strain vs. length of the beam and
Suy ðj!Þ
Gðj!Þ ¼ ð14Þ showed that strain is maximum at the cantilever end of
Suu ðj!Þ
the beam. Thus, the best position for the sensor and
where Suy ð j!Þ is the cross spectral density between the actuator is at the cantilever end, where the modes can
frequency rich input, u(t), and the sensor signal, y(t), be detected and controlled simultaneously. Hence, PZT
and Suu ð j!Þ is the auto spectral density of the frequency patches of lead zirconate titanate are surface bonded at
rich input, u(t). An average frequency response is the cantilevered end of the beam. One PZT (model no.
obtained by segmenting the time signals, calculating QP 10S) as a sensor and two PZT actuators (model no.
Gð j!Þ for each segment, and averaging the Gð j!Þ for QP 10 W) as actuators are surface bonded to the beam.
the FRFs. The effects of truncating the time signal are The PZT QP 10 W and QP 10 S are products from ACX
minimized by using a Hanning window on each seg- Incorporation. The properties of PZT are shown in
ment. In the uncompensated system, i.e. Gc ð j!Þ ¼ 1, Table 2.
the open loop system uncompensated frequency To actuate the PZT actuators, an amplifier, which
response Lð j!Þ is the same as the plant frequency amplifies the input voltage to twenty times of the orig-
response Gð j!Þ. inal one, is used. An anti-aliasing filter is also used to
In Equation (14), the uncertainty parameter a is prevent aliasing. For implementing the controller in
dropped because the measured frequency response is real time, a dSPACE digital data acquisition and real-
used for controller design. Since this is a measured fre- time control system is used.
quency response, the uncertainty related to modeling The dSPACETM uses a DS1103 digital signal process
error does not exist. However, uncertainty due to board for real-time control implementations.
aging, implementation issues such as A/D conversions,
sensor noise, etc., exist. Therefore, the measured FRF
will contain uncertainties that can be displayed on the
sensitivity chart, if known. In the present study the Table 1. Beam properties
robustness considerations are incorporated by enfor-
Symbol Quantity Units Value
cing gain and phase margin constraints via an upper
bound on sensitivity. L Beam Length mm 736.5
wb Beam width mm 53.1
tb Beam thickness mm 1
4. Application to vibration control
b Beam density kg/m3 2690
of a flexible beam E Modulus of elasticity N/m2 7.03  1010
4.1. Experimental setup
The proposed feedback control methodology was
implemented on a flexible beam structure with PZT
sensors and actuators. The flexible beam used in this
experiment is made of aluminum and its properties are

Control Input Sensor Output


(V) (V)
G(s, a)
U(s) Y(s, a)

Figure 5. Open loop uncompensated system. Figure 6. Experimental setup.

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


2000 Journal of Vibration and Control 17(13)

Table 2. PZT patch properties

Symbol Quantity Units PZT actuator PZT sensor

Lwt Dimensions mm 46  33.27  0.25 14  7  0.25


d33 Strain coefficient C/N 7.41  1010 7.41  1010
d31 Strain coefficient C/N 2.74  1010 2.74  1010
rp PZT density kg/m3 7500 7500
Ep Young’s modulus N/m2 6.3  1010 6.3  1010

Input sweep signal 50


0.4 Sweep Sine

Magnitude (dB)
Schroder wave
Input signal (Volts)

0.2 0

0
–50
–0.2
–100
–0.4 101 102
0 10 20 30 40 50 60 70 80 90 100
Output sensor signal 0
1
Phase(degrees)
–500
Sensor (volts)

0.5

0
–1000
–0.5
–1500
–1 101 102
0 10 20 30 40 50 60 70 80 90 100 Frequency (rad/sec)
Time (sec)

Figure 8. Identified open loop uncompensated transfer func-


Figure 7. Time response of the sweep sinusoidal input and
tion. (—)Sweep sine excitation, () Schroeder wave excitation.
measured output signal.

The dSPACETM system also has integrated analog- The time response for the second test is not shown
to-digital and digital-to-analog converters. here due to its similarity to that shown in Figure 7.
Transfer functions identified using the above fre-
quency rich signals are plotted in Figure 8.
The FRFs are calculated using a Hanning window
4.2. Nonparametric system identification
and a 75% overlap. As seen from this figure, both the
Experiments are conducted to identify the FRF of the transfer function estimates have similar magnitude
system. A frequency rich signal is applied as an input to characteristics and phase characteristics. The first
the plant to identify its FRF. Two tests with frequency three target modes of control viz. 1.64 Hz (10.30 rad/
rich inputs, i.e., the sweep sine excitation and the sec), 10 Hz (62.8 rad/sec) and 27.13 Hz (170.48 rad/sec)
Schroeder wave excitation, are conducted for evalua- are clearly seen from the FRF plot.
tion and validation. The Schroeder wave is used for The uncompensated open loop transfer function of
identification purposes because it has a flat auto spec- the beam identified using sweep sine excitation is plot-
trum in the specified frequency range, thereby a more ted on the sensitivity chart in Figure 9. The 0 dB
likely chance of excitation of modes. Sensitivity contour appears as the inverted U in the
The first evaluation test was to excite the beam with sensitivity chart. The three resonance frequencies are
a sweep sine signal with a frequency of 0–50 Hz in a indicated by the ‘«’ symbol on the open loop uncom-
time period of 100 seconds. The second validation test pensated frequency response. By using the sensitivity
was with a Schroeder wave excitation of 0.1–45 Hz for a chart, one can get an idea of how much sensitivity
period of 25 seconds, although data was collected for reduction can be achieved using feedback. Say for
30 seconds so as to measure the vibrations of the first instance, if Gc ðsÞ ¼ 1 in Figure 2, i.e., no controller
mode which remained during the last few seconds of the is employed and feedback is done, there will be
excitation period. The time response signals for sweep approximately 1 dB amplification in the first mode
sine wave excitation are plotted in Figure 7. of 1.64 Hz, 6 dB reduction in the second mode of

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 2001

40 –40 db Frequency response of controller


60
30
40 Gc(10j)

Magnitude (dB)
20 –20 db
20 Gc(62.8j)
L(170.48j)
10 L(62.8j) –12 db
0
Open–Loop Gain (dB)

0 –6 db Gc(170.5j)
6 db –20
–3 db
–10 3 db
L(10.3j) –40
–20 1 db –1 db 100 101 102
0.5 db
–30 0.25 db 0

–40 –200

Phase(degrees)
–50 –400

–60 –600

–70 –800
–1400 –1200 –1000 –800 –600 –400 –200 0
0 db –1000
Open–Loop Phase (deg) 100 101 102
Frequency (rad/sec)

Figure 9. Open loop uncompensated frequency response on


extended sensitivity chart. Figure 10. Frequency response of the controller.

10 Hz, and 8 dB reduction in the third mode of attenuation region, a phase lag of about 800 degrees
27.13 Hz. is provided. This is accomplished by introducing a
second order pair of non-minimum zero and a stable
pole to control the third mode of 170 rad/sec. In par-
4.3. Design of controller Gc(s) ticular, the non-minimum complex zero is at a fre-
For robustness considerations in the control design, the quency of 120 rad/sec with the damping ratio of
interior of the þ6 dB contour is chosen as a forbidden z ¼ 0:5, and the stable complex poles are at a fre-
region, implying that the open loop frequency response quency of 155rad/sec with a damping ratio of
cannot be contained within this region. This choice pro- p ¼ 0:05. Finally, the controller DC gain is adjusted
vides a reasonable magnitude of 6 dB and a phase until the open loop frequency response of Lð j!Þ borders
margin of 30 for stability and a limit on sensitivity the 6 dB contour. The complex poles introduced at
amplification for performance. The control design 4.5 rad/sec also roll off the open loop gain at higher
objective is to place the three selected resonance fre- frequencies to attenuate sensor noise. The overall con-
quencies in the sensitivity attenuation region on the troller is given by Equation (15), and its frequency
extended sensitivity chart. This is achieved by giving a response is shown in Figure 10. The open loop fre-
proper phase at the designated frequencies. The DC quency response of the compensated system is shown
gain is adjusted until the magnitude borders the in Figure 11. From Figure 11, an approximate 8 dB
þ6 dB contour. attenuation in the first mode and a 12–14 dB attenua-
From Figure 9, the frequency at 10 rad/sec is in the tion in both the second and third resonant frequencies
sensitivity amplification zone. If 25 dB of amplification for the closed loop system are anticipated.
and a phase lag of about 170 degrees are provided at the  
frequency of 10 rad/sec, then this mode will be moved to 40
Gc ðsÞ ¼ 1:5
the sensitivity attenuation region. This is achieved by ðs=4:5Þ2 þ 2  0:25ðs=4:5Þ þ 1
introducing a second order pole at 4.5 rad/sec and with  
ðs=35Þ2  2  0:3ðs=35Þ þ 1
a damping ratio of 0.25. Adding 5 dB of gain and of 500 
phase lag at 62.8 rad/sec would also place the open loop ðs=55Þ2 þ 2  0:2ðs=55Þ þ 1
 
frequency response into a lower sensitivity region at that ðs=120Þ2  2  0:5ðs=120Þ þ 1
 ð15Þ
frequency. This is achieved by introducing a second ðs=155Þ2 þ 2  0:05ðs=155Þ þ 1
order pair of non-minimum zero and a stable pole.
Specifically, the non-minimum complex zero is at a fre-
quency of 35 rad/sec with a damping ratio of z ¼ 0:3,
and the stable complex poles are at a frequency of
4.4. Experimental performance validation
55 rad/sec with a damping ratio of p ¼ 0:2.
To control the third mode, amplitude modification The designed controller was implemented on the exper-
is not required because a simple feedback will provide imental setup, and its performance was evaluated. Two
9 dB attenuation. To move this mode in sensitivity tests were performed to evaluate the controller’s

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


2002 Journal of Vibration and Control 17(13)

40 –40 db 0
With loop shaping
30 Without control
–10
20 –20 db
L(62.8j) –20
L(170.48j) L(10.3j)
10 –12 db
Open–Loop Gain (dB)

Energy level in Decibels


0 –6 db –30
6 db –3 db
–10 3 db –40
–20 1 db –1 db
0.5 db –50
–30 0.25 db
–60
–40

–50 –70

–60
–80
–70
–2000 –1500 –1000 –500 0
0 db –90
Open–Loop Phase (deg) 0 20 40 60 80 100
Frequency (Hz)

Figure 11. Open loop compensated frequency response.


Figure 13. Power spectral density comparison plot
(6–15 seconds).

p
die are now suppressed within 3 seconds. The power
1
Without control
spectral density comparisons plots for 6–15 seconds of
With loop shaping the data are shown in Figure 13. The power spectral
0
density from 6–15 seconds is considered because the
–1
(a) sudden kicking of control action after the excitation
5 6 7 8 9 10 11 12 13 14 15
at 5 seconds might lead to excitation of higher modes,
Sensor output (Volts)

1
Without control which die down because of their own damping over the
With PP control
0 next second. Evidently, substantial drops are observed
in the first three modes of the beam.
(b)
–1
5 6 7 8 9 10 11 12 13 14 15
1 4.4.2. Schroeder wave excitation. In the second test,
Without control
With LQR control
the performance of the controller was evaluated when
0 the beam is under the persistent excitation of the
(c) Schroeder wave. Since the beam has two actuators,
–1
5 6 7 8 9 10 11 12 13 14 15 one actuator was used for excitation and the other
Time (sec) was used to control the vibrations. The Schroeder
wave, with a frequency range of 0.5–45 Hz and duration
Figure 12. Time response comparisons for initial condition of 25 seconds, was used for excitation of the beam. The
test (a) Loop shaped control (b) Pole placement control (c) LQR controller was implemented by the real time data acqui-
control. sition system with a sampling frequency of 250 Hz.
The open loop (uncompensated) and closed loop
performance (Initial Condition tests and Schroeder (compensated) time responses are shown in Figure 14.
wave excitation tests). Also, shown in Figure 15 is the transfer function
magnitude plot for the uncompensated and compen-
4.4.1. Initial condition tests. In this test, the beam sated open loop systems. The measured magnitude
structure is excited using three sinusoidal signals of drops at the three resonant frequencies and are, respec-
modal frequencies for the initial 5 seconds at 40 Volts tively, 9 dB for the first mode, 15 dB for the second
each, together with a noise input. The controller was mode, and 14 dB for the third mode. These values are
implemented in the real time data acquisition system close to the predicted values. As mentioned earlier, sen-
with a sampling frequency of 250 Hz. The open loop sitivity attenuation in one region is followed by sensi-
test is conducted to view the results in absence of the tivity amplification in another. This is also visible from
controller, and the free vibrations (open loop uncom- the transfer function plot where at some frequencies
pensated) results are shown in Figure 12(a). Next, the sensitivity amplification is observed.
controller was implemented. The time response of real
time loop shaped control is shown in Figure 12(a). The 4.4.3. Comparison with time domain based control-
uncontrolled vibrations that took about 15 seconds to ler results. The same experimental setup was studied

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 2003

Time response 20
Experimental data
Open loop uncompensated 10 Full 8th order model
0.5 Loop compensated Reduced 6th order model
0.4 0

0.3 –10
Sensor signal (volts)

0.2

Magnitude(dB)
–20
0.1
–30
0

–0.1 –40

–0.2 –50

–0.3
–60
–0.4
–70
–0.5
–80
5 10 15 20 25 30 35 101 102
Time (sec) Frequency (rad/sec)

Figure 14. Time response sensor output comparison for Figure 16. Magnitude plot for parametric model identification
Schroeder wave test. (o) Experimental data, (—) 8th Order model, () 6th Order
model.

Transfer function
10
Open loop uncompensated 0
0 Loop compensated Experimental data
Full 8th order model
Reduced 6th order model
–10 –200

–20
–400
Magnitude (dB)

Angle (degrees)

–30

–40 –600

–50
–800
–60

–1000
–70

–80
100 101 102 –1200
101 102
Frequency (rad/sec)
Frequency (rad/sec)

Figure 15. Magnitude plot for the Schroeder wave test. Figure 17. Phase plot for parametric model identification (o)
Experimental data, (—) 8th Order model, () 6th Order
for pole placement and LQR control designs. To pro- model.
ceed with this control design, a parametric model is
needed. In order to obtain the parametric model, an
8th order model that closely matches the transfer func- Figure 12(c), respectively. A comparison of the power
tion magnitude and the phase plot of the system, was spectral density plots for all the three controllers is plot-
identified. However this 8th order model is neither con- ted in Figure 18. All the three controllers achieve sim-
trollable nor observable; hence, it was necessary to con- ilar performance, as seen in Figure 18. The control
duct a model reduction. The reduced 6th order model, effort of the three controllers - LQR, pole placement,
that follows the magnitude and phase of the identified and loop shaped controller - are plotted in Figure 19.
model, was derived. The identification results are plot- Evidently, all the three controllers experienced satura-
ted in Figure 16 and 17. tion in the initial period and then finally decay. The
Two controllers viz. pole placement and LQR effort provided can be quantified in terms of energy
were investigated for multimodal control for the ini- or the 2 norm of the control signal. Table 3 compares
tial condition tests. The time response of pole place- the control energy input to the system for vibration
ment and LQR controller implementation for suppression. With the achievement of a similar control
initial conditions are shown in Figure 12(b) and performance for the flexible beam, Table 3 clearly

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


2004 Journal of Vibration and Control 17(13)

0
Table 3. Control effort comparison
Without control
–10 With loop shaping Controller Control Energy (V2) juj2 2–Norm
With PP control
With LQR control
–20 Loop shaping 44.30 6.65
Pole placement 48.40 6.95
Energy level in decibels

–30
LQR 65.05 8.06
–40

–50
controller can be designed by placing the three
–60
resonance frequencies in the sensitivity attenuation
–70 region on the extended sensitivity chart. The attenua-
tion in the resonant frequencies can be accurately pre-
–80
dicted from the open loop frequency response of the
–90
0 10 20 30 40 50 60
plant and the controller from the extended sensitivity
Frequency (Hz) chart. Approximately, the same attenuation, as pre-
dicted, was achieved for closed loop controller
Figure 18. Power spectral density comparison plots for all the implementation.
three controllers (6–15 seconds). Finally, the results of the loop shaped controller are
compared with the pole placement and LQR control-
lers. With the achievement of a similar control perfor-
0.8
mance for the flexible beam, Table 3 shows that the
With LQR control control energy required in loop shaping control is
With PP control
0.6
With loop shaping
lower than that of the pole placement and LQR con-
trol. One may argue that the control effort of the pole
0.4
placement and LQR controller can be improved by
Control signal (volts)

0.2 trying (iteratively) various pole locations and weighting


matrices.
0
However, in order to implement the pole placement
–0.2 and LQR controller, an accurate analytical model is
required. To obtain the accurate analytical model is
–0.4 often very time consuming.
–0.6
There are always assumptions and complexities
involved in identifying the parametric model of the
–0.8
5 10 15 system. For pole placement and LQR controller, one
Time (sec) also needs an estimator design which is obtained from
the analytical model and thus adding to the complexity
Figure 19. Control Signal comparison plots for three of design and implementation of the controller in real
controllers. time. For loop shaping control, the feedback is the
output from the sensor and is irrespective of the
shows that the control energy required in loop shaping states as required in the time domain based controller.
control is lower than that of the pole placement and In this case, there exists limitations that both the pole
LQR control. placement and LQR control have, in which it is impos-
sible to design since the system is neither controllable
nor observable at such a higher order model for the
5. Conclusions
flexible beam to include the suppression of the fourth
In this paper, the loop shaping method was discussed or higher resonant frequencies. Loop shaping control
and implemented on a flexible beam structure for vibra- clearly offers a winning alternative where the experi-
tion control. Sensitivity charts that relate the open loop mental data is used for controller designs and can be
gain and phase of the plant to the closed loop sensitivity readily extended to other higher modes of the system.
were discussed. The design trade off is also shown
where sensitivity attenuation in one region is accompa- Acknowledgements
nied by sensitivity amplification in another. To avoid The authors would like to thank the support provided by
the effort and complexities of deriving an accurate ana- NSF via two research grants (no. 0305027 and no. 0717860)
lytical model, a model of the plant based on experimen- in conducting this research. This research is also partially
tal system identification was used. It was shown that a supported by an internal grant from the Institute of Space

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


Sethi et al. 2005

System Operations (ISSO) at University of Houston. Any Systems, Man and Cybernetics, Vancouver, BC, Canada,
opinions, findings, conclusions, or recommendations, in this 22–25 October 1995, Vol. 1, pp.557–562.
material are expressed by the author(s) and do not necessarily Gouatarbes J, Boukhnifer M, Ferreira A, Aubry D, and
reflect the views of the sponsor. Magnan P (2007) Robust control for ultrasonic motor
operating within harsh environments. In: Proceedings of
IEEE/ASME International Conference on Advanced
References Intelligent Mechatronics, AIM, Zurich, Switzerland, 4–7
Ahuja MS, Li J-Z, Song G and Franchek MA (2008) September 2007.
Increasing noise transmission loss across an aluminum Gu H and Song G (2005) Active vibration suppression of a
panel using PZT actuators. In: Earth and Space composite I-beam using fuzzy positive position control.
Conference 2008: Proceedings of the 11th Aerospace Smart Materials and Structures 14(4): 540–547.
Division International Conference on Engineering, Science, Gu H and Song G (2007) Active vibration suppression of a
flexible beam with piezoceramic patches using robust
Construction, and Operations in Challenging Environments,
model reference control. Smart Materials and Structures
Long Beach, 3 March 2008, Vol. 323.
16(4): 1453–1459.
Albert DS (1999) Active control of transmission loss in lightly
Hagood NW and Anderson EH (1992) Simultaneous sensing
damped panels. MSME Thesis, Purdue University, IN.
and actuation using piezoelectric materials. Proceedings of
Albert DS, Franchek MA and Bernhard RJ (2000) Active
SPIE 1543: 409–421.
control of transmission loss in lightly damped panels.
Hamer AJD, Angelis GZ, Van de Molengraft MJG, and
Journal of Noise Control Engineering 48(2): 48–59.
Steinbuch M (2005) A practical loop-shaping approach
Bailey T and Hubbard Jr EJ (1985) Distributed piezoelectric
for pole placement in mechatronic systems. In:
– polymer active vibration control of a cantilever beam.
Proceedings of Conference of Control Applications,
Journal of Guidance, Control, and Dynamics 8(5): 605–611.
Toronto, Canada, 28–31 August, pp.394–399.
Bayon de Noyer MP and Hanagud S (1998) Single actuator
Han JH, Tani J and Lee I (1999) Multi-modal vibration con-
and multi-mode acceleration feedback control. Journal of
trol of smart composite plates. Materials Science Research
Intelligent Materials Systems and Structures 9(7): 522–533.
International 5(2): 122–127.
Blanguernon A, Lene F and Bernadou M (1999) Active con-
Hsien TL, Sun YY, Tsai MC and Huang SJ (1997) Robust
trol of a beam using a piezoceramics element. Smart
servo control system design for two flexibly linked masses.
Materials and Structures 8: 116–124. Control Engineering Practice 5(7): 951–961.
Bode HW (1945) Network Analysis and Feedback Amplifier Kaitwanidvilai S and Parnichkun M (2004) Position control
Design. New York: Van Nostrand. of a pneumatic servo system by genetic algorithm based
Butler R and Rao V (1996) A state space modeling and con- fixed-structure robust H infinity loop shaping control. In:
trol method for multivariable smart structural systems. Proceedings of IEEE Industrial Electronics Society, Busan,
Smart Materials and Structures 5: 386–399. Korea, 2–6 November 2004, pp.2246–2251.
Chopra I (1996) Review of current status of smart struc- Kang YK, Kim MK, Park HC and Agrawal B (1999) Multi-
tures and integrated systems. Proceedings of SPIE 2717: modal vibration control of laminated composite plates
20–62. using piezoceramic sensors/actuators. Journal of
Corr LR and Clark WW (2003) A novel semi-active multi- Intelligent Materials Systems and Structures 9(12):
modal vibration control law for a piezoceramic actuator. 988–990.
Journal of Vibration and Acoustics 125(2): 214–222. Kar IN, Seto K and Doi F (2000) Multimode vibration con-
Crawley EF and de Luis J (1987) Use of piezoelectric actua- trol of a flexible structure using H1 -based robust control.
tors as elements of intelligent structures. AIAA Journal IEEE/ASME Transactions on Mechatronics 5(1): 23–31.
25(10): 1373–1385. Kim JH, Choi SB, Cheong CC, Han SS and Lee JK (1999)
Den Hamer AJ, Angelis GZ and Roozen NB (2005) Broad- H1 control of structure-borne noise of a plate featuring
band active vibration suppression using PPF focused on piezoceramic actuators. Smart Materials and Structures
industrial application. IEEE/ASME Transactions on 8(1): 1–12.
Mechatronics 10(2): 146–153. Lazarus KB and Crawley EF (1996) Multivariable high-
Ding J, Wu S-C and Tomizuka M (2004) Settling control with authority control of plate-like active structures. Journal
reference redesign for dual actuator hard disk drive sys- of Guidance, Control, and Dynamics 19(6): 1357–1363.
tems. Annual Reviews in Control 28(2): 219–227. Li CJ and Ulsoy AG (1999) High precision measurement of
Falangas ET, Dworak JA and Koshigoe S (1994) Controlling tool-tip displacement using strain gauges in precision flex-
plate vibrations using piezoelectric actuators. IEEE ible line boring. Mechanical Systems and Signal Processing
Control Systems Magazine 14(4): 34–41. 13: 531–546.
Folcher J-P (2002) Robust design framework for an Manning WJ, Plummer AR and Levesley MC (2000)
underwater vehicle motion controller. In: Oceans ’02 Vibration Control of a flexible beam with integrated actu-
MTS/IEEE, Biloxi, Mississippi, 29–31 October 2002, ators and sensors. Smart Materials and Structures 9:
pp.173–179. 932–939.
Fujimoto T (1995) Vibration control of a cantilever beam McFarlane D and Glover K (1992) A loop-shaping design
using a H infinity controller for noncollocated systems. procedure using H1 synthesis. IEEE Transactions on
In: Proceedings of the IEEE International Conference on Automatic Control 37(6): 759–769.

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012


2006 Journal of Vibration and Control 17(13)

Oueini SS and Nayfeh AH (1997) Multimode control of Journal of Intelligent Materials Systems and Structures
flexible structures using saturation. Collection 13(1): 13–22.
of Technical Papers - AIAA/ASME/ASCE/AHS/ASC Sethi V and Song G (2004) Multimode optimal vibration
Structures, Structural Dynamics & Materials Conference control of flexible structure using piezoceramics. In:
1: 597–602. Proceedings of IEEE Conference on Control Applications,
Park K, Kim S, Choi D and Sohn B (2008) An active vibra- Taiwan, 2–4 September, pp.216–221.
tion isolation system using a loop shaping control tech- Sethi V, Song G and Franchek MA (2008) Loop shaping
nique. In: Mechatronics and Embedded Systems and control of a model-story building using smart materials.
Applications, Beijing, China, 12–15 October 2008, Journal of Intelligent Material Systems and Structures 19:
586–590. 765–777.
Prakah-Asante KO and Craig KC (1994) The application of Song G and Gu H (2007) Active vibration suppression
multi-channel design methods for vibration control of an of a smart flexible beam using a sliding mode based
active structure. Smart Materials and Structures 3: controller. Journal of Vibration and Control 13(8):
329–343. 1095–1107.
Rew KH, Han JH and Lee I (2002) Multi-modal vibra-
tion control using adaptive positive position feedback.

Downloaded from jvc.sagepub.com at UNIV TORONTO on November 11, 2012

Вам также может понравиться