Você está na página 1de 11

Home Search Collections Journals About Contact us My IOPscience

Comparison of negative and positive position feedback control of a flexible structure

This content has been downloaded from IOPscience. Please scroll down to see the full text.

2011 Smart Mater. Struct. 20 015011

(http://iopscience.iop.org/0964-1726/20/1/015011)

View the table of contents for this issue, or go to the journal homepage for more

Download details:

IP Address: 148.247.124.101
This content was downloaded on 02/08/2017 at 19:07

Please note that terms and conditions apply.

You may also be interested in:

Optimal and robust modal control of a flexible structure using an active dynamic vibrationabsorber
Sang-Myeong Kim, Semyung Wang and Michael J Brennan

Robust broadband vibration control of a flexible structure using an electrical dynamicabsorber


Sang-Myeong Kim, Semyung Wang and Michael J Brennan

Active vibration control using delayed resonant feedback


Sang-Myeong Kim and Michael J Brennan

Feedback control laws for proof-mass electrodynamic actuators


C González Díaz and P Gardonio

Negative derivative feedback for vibration control of flexible structures


G Cazzulani, F Resta, F Ripamonti et al.

Demonstration of non-collocated vibration control of a flexible manipulator using electrical


dynamic absorbers
Sang-Myeong Kim, Heungseob Kim, Kwangsuck Boo et al.

Downscaling of proof mass electrodynamic actuators for decentralized velocity feedbackcontrol on a


panel
Paolo Gardonio and Cristóbal González Díaz

Hybrid PD and effective multi-mode positive position feedback control for slewing and vibration
suppression of a smart flexible manipulator
Jun-qiang Lou, Yan-ding Wei, Yi-ling Yang et al.
IOP PUBLISHING SMART MATERIALS AND STRUCTURES
Smart Mater. Struct. 20 (2011) 015011 (10pp) doi:10.1088/0964-1726/20/1/015011

Comparison of negative and positive


position feedback control of a flexible
structure
Sang-Myeong Kim1 , Semyung Wang1,3 and Michael J Brennan2
1
Mechatronics, Gwangju Institute of Science and Technology, 261 Cheomdan-gwagiro,
Buk-gu, Gwangju, 500-712, Korea
2
Institute of Sound and Vibration Research, University of Southampton, University Road,
Highfield, Southampton SO17 1BJ, UK

E-mail: smkim123@hanmail.net, smwang@gist.ac.kr and mjb@isvr.soton.ac.uk

Received 24 August 2010, in final form 8 November 2010


Published 9 December 2010
Online at stacks.iop.org/SMS/20/015011

Abstract
Two modal control techniques, negative position feedback (NPF) and positive position feedback
(PPF), are applied to reduce multi-mode vibration of a lightly damped flexible beam using a
piezoelectric sensor and piezoelectric actuators. The NPF and PPF controllers are constructed
by respectively feeding back the generalized displacement response from the sensor in a
negative and a positive sense to the actuators through second order high pass and low pass
filters. PPF is well known while NPF is new for this application and is in fact an electrical
realization of a dynamic vibration absorber. The choice of the parameters for controllers of both
types is made easy by a robust modal control technique that offers an optimal performance for
NPF control and a near-optimal performance for PPF control. Explicit forms of the controller
parameters are presented. Experiments are conducted on a cantilever beam embedded with a
matched pair of PZT (lead zirconate titanate) patch actuators and a collocated PVDF
(polyvinylidene fluoride) patch sensor. The experiments demonstrate that it is possible to
realize an electrical dynamic absorber using the generalized displacement sensor. It is further
demonstrated that NPF can be a good alternative control strategy particularly when multiple
modes are to be controlled.
(Some figures in this article are in colour only in the electronic version)

1. Introduction mode. There are also some difficulties in PPF control. For
example, it has an inherent static stability problem at 0 Hz [1].
Positive position feedback (PPF) has been widely used to The controller is also sensitive to the dynamics of the modes
reduce the vibration of a flexible structure which has a of lower order than the target mode, and little information is
large number of modes [1]. PPF was originally developed available regarding simple and explicit rules for the design of
to overcome the practical shortcomings when implementing the controller parameters. These issues are addressed in this
direct velocity feedback (DVFB) to this application [2], such paper.
as transducer installation problems and finite bandwidths of A technique called negative position feedback (NPF),
transducers [3]. For the former, PPF control systems often use which has been presented under a different name [4] and was
piezoelectric transducers that can be easily embedded within briefly discussed in [5], is employed here to overcome the
the structure. For the latter, PPF control uses a low pass difficulties mentioned above. An optimal robust control theory
filter such that it is insensitive to the high frequency transducer based on the Nyquist stability criterion [4] is used. There
dynamics and also to the modes of higher order than the target are only subtle differences between NPF and PPF control
in terms of construction. The NPF (or PPF) controller is
3 Author to whom any correspondence should be addressed.
constructed by feeding back the negative (positive) generalized

0964-1726/11/015011+10$33.00 1 © 2011 IOP Publishing Ltd Printed in the UK & the USA
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

The two modal controllers considered in this paper are



M
HNPF (jω) = ka,n (jω)2 a−,n1 , (2)
n=1


M
HPPF(jω) = − ka,n ωa2,n a−,n1 , (3)
n=1

Figure 1. A smart cantilever beam embedded with a matched pair of


where a,n = [ωa2,n − ω2 + jbn ωa,n ω], n follows the structural
PZT patch actuators (one on top and the other on bottom of the mode number from 1 to M, and the subscript a indicates an
beam) and a collocated PVDF patch sensor, subject to a primary absorber since both ‘resonant’ controllers behave like dynamic
excitation f p . vibration absorbers in a mechanical sense. Thus each filter
has the three parameters ωa,n , bn, and ka,n , and is used to
control the n th structural mode. Equation (2) describes the
displacement response from the sensor to the actuator through NPF controller which has a second order high pass filter while
a second order high pass (low pass) filter. It is shown here equation (3) describes the PPF controller which has a second
that the NPF controller is an electrical realization of a dynamic order low pass filter. The minus sign in equation (3) within
vibration absorber [6] if a generalized displacement sensor is the negative feedback control configuration, −H (jω), means
used. Explicit forms of the controller parameters of NPF and that it is positive feedback. It should be noted that the two
filters are similar in the ‘resonant’ regions (substitute ωa,n into
PPF control are presented which offer an optimal performance
equation (2)) and so will be the resulting performances. The
for NPF control and a near-optimal performance for PPF
tuned
 natural frequency of the n th absorber filter is ωa,n =
control. By comparing the two modal control techniques, the
ka,n /m a,n in which ka,n is the equivalent spring coefficient
aim of this paper is to illustrate and contrast the performance
and m a,n is the equivalent mass of the filter, bn = 2ζa,n is the
and robustness characteristics of NPF and PPF control for a normalized half-power bandwidth with ζa,n being the damping
distributed parameter system. ratio, and ka,n can also be interpreted as the filter gain from
a control point of view. The structural and absorber modal
damping ratios are defined as
2. Theory 
ςs,m = cs,m /(2 m s,m ks,m ), (4a )

2.1. Two modal controllers ςa,n = m a,n ka,n /(2ca,n ). (4b )
Equation (4a ) also relates to an R –C – L series electrical circuit
Consider vibration control of a cantilever beam embedded with
in which R is a resistor, C is a capacitor, and L is an inductor
a matched pair of PZT patch actuators (one on each side of the
(equivalently, an m  c  k parallel oscillator in which m is
beam) [7] and a collocated PVDF patch sensor as shown in
mass, c is a damper, and k is a spring), while equation (4b )
figure 1. The beam is subject to an arbitrary primary excitation
relates to an R  C  L parallel circuit [8] (equivalently, m –c–
f p , the control output is f c , and −H (jω) indicates that is a k series oscillator. see appendix A). The static stiffness of the
negative feedback controller. An accelerometer is attached to plant is given by ks = (ks−,11 + ks−,21 + · · · + ks−,res
1 −1
) , whereas
monitor the end tip vibration before and after control. The term that of the filter is ka = (ka,1 + ka,2 + · · ·).
‘collocation’ is used with the assumption that the beam acts as Multiplying the receptance-like expression (the ratio of
a one-dimensional structure. It is assumed that the actuators the generalized position or displacement to excitation) in
generate a pure bending moment and the plant response (i.e., equation (1) by jω gives the generalized structural mobility,
the sensor output to the actuator input) within the frequency and dividing equation (2) by jω gives the generalized absorber
range of interest is written as impedance (force divided by velocity). It is clear from
the generalized mobility that the plant is a minimum phase

M system such that the phase angle  is −90◦   jωG(jω) 
G(jω) = ks−,m1 ωs2,m s−,m1 , (1) 90◦ [9]. It should be noted that the structural mobility and
m=1
the absorber impedance exactly match those for the nominal
√ plant and the electrical dynamic vibration absorber for an
where s,m = [ωs2,m − ω2 + jam ωs,m ω] in which j = −1, active vibration isolation system [4] that uses either a velocity
the subscript m denotes the structural mode number from 1 or an acceleration sensor. Thus, the NPF controller is in
to M inwhich M is the total number of modeled modes, fact an electrical realization of the dynamic vibration absorber
ωs,m = ks,m /m s,m is the m th structural natural frequency in that uses a generalized displacement sensor. Details of such
which ks,m is the modal stiffness and m s,m is the modal mass, equivalences are summarized in appendix A for a single
and am = 2ζs,m is the normalized half-power bandwidth of the degree of freedom (DOF) oscillator plant. A mechanical
m th mode, in which ζs,m is the modal damping ratio. A residual analogous model is also presented. Thus NPF is a passivity-
stiffness term ks−,res
1
can be further added to equation (1), if based controller and is unconditionally stable [4], provided
required, to incorporate the effect of the un-modeled higher equations (1)–(3) are valid for all frequencies. Such an analogy
frequency modes. is not possible for PPF.

2
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

2.2. Cost function and the constraint When the structural mode under consideration is the TM
and the control filter is perfectly tuned such that β = 1, this is
The vibration reduction ratio S(jω) = e/d (or the sensitivity
an auto-coupling and equations (7) and (8) reduce to
function), where e and d are the plant responses after
and before control, respectively, is adopted as the control L NPF (0) → −κnn 2 , (9a )
performance. It can be written in decibels as
L NPF ( j ) = gnn , (9b )
RR (dB) = 20 log10 |S(jω)|, (5) L NPF (∞) → −κnn  , −2
(9c)
where S(jω) = [1 + L(jω)]−1 in which L(jω) = G(jω)H (jω) L PPF (0) → −κnn , (10a )
is the open-loop frequency response function (FRF) of the L PPF ( j ) = gnn , (10b )
control system. In this paper, the control system is defined to −4
be stable and robust with a degree of r if and only if its open- L PPF (∞) → −κnn  , (10c)
loop FRF locus does not enclose or cross the circle of radius where the subscript nn is now used instead, and the values
r centered at the instability point (−1, 0) [4]. This robustness for L(0) and L(∞) are asymptotes when  → 0 and when
condition can be written down as  → ∞, respectively. In general, for control performance it
is desirable that the loop gain should be large and real- and
RR (dB)  (−20 log10 r ), (at all frequencies) (6) positive-valued in the effective region, while it is desirable
for robustness that the loop gain should be small in the
where the limiting value on the right is the maximum allowable ineffective regions [4]. In this context, both controllers work
control spillover. Robust optimal control is to achieve the effectively at the resonance frequency when  = 1; in fact they
best performance (i.e., minimize RR) without violating the behave like a skyhook damper of damping coefficient ca,n =
constraint in equation (6). gnn cs,n , yielding an active modal damping ratio of ςs,n(active) =
gnn ςs,n [4]. The asymptotes indicate that NPF control is
2.3. Coupling analysis insensitive at both low and high frequencies compared to the
TM (by an order of −2 ). PPF control is further insensitive
Consider the case when a single (absorber) filter is used in
at frequencies greater than the TM (by an order of −4 ),
the control of one of a large number of structural modes
however, it is sensitive at frequencies lower than the TM.
of a flexible structure. The structural mode is termed the
NPF is unconditionally stable, while the PPF is not and the
target mode (TM), the other uncontrolled modes are called
Nyquist stability condition dictates that the gain should be
the mode(s) of concern (MOC), and the most critical MOC
within 0 < κnn < 1 in equation (10a ) [1]. Thus the amount of
is the mode of critical concern (MOCC) from the point of
reduction given in equation (10b ) has to be limited even when
view of stability and robustness. The coupling between the
an ideal single DOF oscillator plant is considered. When the
filter and the TM is called auto-coupling while the coupling
structural mode under consideration is again the TM but now
between the filter and an MOC is called cross-coupling. The
the filter frequency tuning ratio is β = 1, equation (10) at
coupling/loading among either structural modes or control
 = 1 reduces to
filters is called mutual coupling/loading. It is assumed that L NPF (j) ≈ gnn , (11a )
the plant is lightly damped and the modes are well separated
such that the response in each resonance region is governed by L PPF (j) ≈ gnn β 2 , (11b )
−1
a single mode (i.e., the mutual loadings are negligible). provided that |x nm |/bn 1 in which x nm = β − β indicates
The open-loop gain indicates the control effort as well a normalized mistuning bandwidth. Although there is a subtle
as the strength of coupling between the structure and the change in the PPF performance according to the tuning ratio,
absorber [4, 10]. The open-loop FRF for the coupling between both work well in the resonant region as both are real- and
the m th structural mode and the n th absorber filter can be positive-valued.
written as When the structural mode is now an MOC with αnm = 1,
L NPF (j) = −κmn αnm
2
2 (1 − 2 + jam )−1 at  = 1 with the condition |x nm |/bn
1 in which x nm =
−1
αnm − αnm equation (10) reduces to
× (1 − αnm
2
2 + jbn αnm )−1 , (7)
−1 −1
L PPF (j) = −κmn (1 − 2 + jam )−1 L NPF (j) = jgnn bn δnm x nm , (12a )
× (1 − αnm
2
2 + jbn αnm )−1 , (8) −1 −1 −2
L PPF (j) = jgnn bn δnm x nm αnm , (12b )
where  = ω/ωs,m is a dimensionless frequency normalized where −1
gnn δnm = gmn in which δnm = cs,m /cs,n . It can be
−2
by ωs,m , κmn = ka,n /ks,m = σmn αnm is the effective spring seen that, regardless of whether it is a higher or a lower MOC,
ratio, σmn = m a,n /m s,m = gmn am bn αnm is the effective the loop gain of NPF control diminishes as the MOC is farther
mass ratio, and gmn = ca,n /cs,m is the ratio of damping away from the TM. The loop gain of PPF control additionally
coefficients. The non-dimensional variable αnm = ωs,m /ωa,n depends on αnm−2
, indicating that it diminishes more rapidly for
can be interpreted as either the inverse of the frequency tuning a higher order MOC but unfortunately increases for a lower
ratio when the mode is the TM (i.e., auto-coupling) or the order mode when αnm 1. This undesirable sensitivity in the
natural frequency ratio when the mode is a MOC (i.e., cross- low frequency region is a reason of introducing NPF control in
−1
coupling). For the former, the notation β = αnm is used. this paper.

3
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

2.4. Design rules issue, which is the static stability at 0 Hz. Applying the same
robust constraint of 2 dB control spillover to this case gives
In this section, the three parameters of both NPF and PPF
control filters (ωa,n , bn , and ka,n ) are determined analytically. ka,n = ks (1 − ro ), (17b )
The design method is developed for NPF control and is then
simply extended to PPF, as the PPF controller in equation (3) where ks = (ks−,11 + ks−,21 + · · · + ks−,res
1 −1
) is the static stiffness of
behaves similarly to the NPF controller in equation (2) in the the plant. The smaller of equations (17a ) and (17b ) is chosen
resonant region. The method offers an optimal performance as the robust gain in this paper. This gain is then substituted
for NPF control while it offers a near-optimal performance for into equation (14) to obtain the optimal filter bandwidth.
PPF control.
When a single NPF filter is used to control single mode 2.5. Examples
(TM) without considering any other modes (MOC), applying
Den Hartog’s fixed-point theory for vibration absorbers [6] 2.5.1. Single DOF structure. The NPF and PPF controllers
with the aim of achieving the maximally flat mobility described in section 2.1 were applied to a single DOF oscillator
frequency response function (FRF) gives the optimal tuning with a damping ratio of ςs = 0.01. As there is no limit
frequency and the optimal bandwidth given by of the NPF gain that can be imposed in this case, the PPF
gain from equation (17b ) was used for both controllers for
ωa,n = ωs,n , (13) the purpose of comparison. Figures 2(a)–(d) show open-loop
 FRFs (a), reduction ratios (b), displacement (c) and velocity
bn = 2κnn , (14) (d) responses before (dotted lines) and after NPF (solid) or PPF
where κnn = ka,n /ks,n in which ka,n and ks,n are assumed to (dashed) control. The 2 dB robustness boundary circle and line
be given, and the details of derivation is given in appendix B. (dash-dot) are also indicated in (a) and (b), respectively. As
These rules also induce a maximum control spillover of intended, the NPF control offers the maximally flat mobility
about 2 dB within the auto-coupled TM response itself (see FRF in figure 2(d) without violating the maximum control
appendix B). spillover constraint of 2 dB as can be seen in figures 2(a)
If all MOC are now considered and, for convenience, the and (b). Whereas, the PPF controller designed to meet the
same spillover constraint is applied also to all cross-coupled static robustness (note the loop (dashed) starting point at
cases, the complete system is subject to a maximum control (−1 + ro , 0) in figure 2(a)) gives slightly larger and smaller
spillover of 2 dB at all frequencies. This action imposes a control spillovers just below and above the resonance region
certain restriction on the maximum loop gain of the MOC in figure 2(b), respectively. Nevertheless, the PPF controller
given in equations (12a ) and (12b ), from which the last performs as well as the NPF controller in the resonance region
unknown parameter, the filter gain ka,n , is then determined. in figures 2(c) and (d). Further, it exhibits a near-optimal
This statement can be written down mathematically as performance yielding a quite flat displacement FRF as can be
seen in figure 2(c).
|L(j)|  lMOC , (15)
2.5.2. Two DOF structure. In flexible structures, it is very
where lMOC is the maximum allowable loop gain of the MOC
likely that, when the TM is the fundamental mode, the MOCC
and can be written as lMOC = (ro−1 − ro ) in which ro = 10−2/20
for NPF control is the second mode while the robustness
because of the 2 dB spillover constraint and the equation can
concern using PPF control occurs due to static instability at
be obtained from the right triangle having corners at (−1, 0),
0 Hz. It is thus interesting to examine the condition when the
(0, 0), and (0, ±jlMOC /2) in the Nyquist plot. Since κnn =
two controllers has the same gain. Equating equations (16)
gnn an bn , combining equations (12a ), (14), and (15) gives the
and (17b ) gives
optimal NPF gain as
δ12 |x 12| = a1−1 (ks /ks,1 )(1 − ro )/lmoc . (18)
ka,n = ks,n an lMOC δnm |x nm |, (16)
When a1 = 0.02, ks ≈ ks,1 , and the 2 dB cross-coupling
where δnm = cs,m /cs,n is related to the relative magnitude
spillover constraint, for example, it yields δ12 |x 12| ≈ 22. If
between the TM and MOC resonance peaks, and |x nm | =
−1 δ12 |x 12| > 22, NPF control performs better.
|αnm − αnm | with αnm = ωs,m /ωs,n is related to the relative
If now the second mode is the TM and the fundamental
distance between the TM and MOC natural frequencies. The
mode is the MOCC, comparing equations (16) and (17a ) shows
smallest value of equation (16) is chosen from all the candidate
that the NPF controller can have a larger gain by a factor of
(M − 1) MOC, and the mode giving the value is the MOCC. −2
α21 . This case is simulated with an arbitrary two DOF system
Similarly, the optimal PPF gain is written as
with parameter values ςs2 = 0.01, δ21 = 1, and α21 = 0.1,
ka,n = ks,n an lMOC δnm |x nm |αnm
2
. (17a ) and the results are shown in figure 3 in a similar format to
that of figure 2. The circle-shaped locus (solid or dashed
Note that the choice of 2 dB spillover constraint for cross- line) in figure 3(a) is for the MOCC whose maximum gain is
coupling is arbitrary and has been used merely as a reference limited by equation (15). Figure 3(b) shows that the design
to derive equations (16) and (17a ), which then hold for any rules yield a maximum control spillover of about 2 dB for both
control spillover. In PPF control, there is a critical stability NPF and PPF control at all frequencies. It is not clearly seen

4
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

Figure 2. Open-loop FRFs (a), reduction ratios (b), displacement (c) and velocity (d) responses of a single DOF structure of ςs = 0.01 before
(dotted) and after NPF (solid lines) or PPF (dashed) control. The 2 dB robustness boundaries (dash-dot) are indicated in (a) and (b).

in figures 3(c) and (d) but the loading of the absorber slightly and a collocated PVDF patch sensor (Measurement Specialties,
pushes the first mode to a lower frequency. If the 2 dB control DT2-028K, 60 mm × 12 mm × 0.028 mm) at its root as
spillover constraint for cross-coupling is completely relaxed as shown in figure 4. Constrained layer damping treatment was
an extreme case, the NPF gain can have any positive value applied to the beam to give moderate passive damping by
but note that the PPF gain has to be limited by the static, using double-sided sticky tape and aluminum tape. To monitor
robust stability given by equation (17b ). If the fundamental the vibration before and after control, an accelerometer (B&K
mode and the second mode are controlled simultaneously with 4393) was additionally attached to the end tip along the center
also relaxing the spillover constraint for cross-coupling, the line of the beam and an impact force was applied to the
total stiffness (gain) of the PPF type absorber is limited by top of the accelerometer using a hammer (Endevco 2302-
ka = ks (1 − ro ) from equation (17b ), where ka = (ka,1 + ka,2 ), 10) shown in figure 4. The detailed feedback loop circuit is
indicating that the total gain must be shared. NPF control is
shown in figure 5, where the charge amplifier K s (NEXUS)
free from such a restriction because it is unconditionally stable.
was used for conditioning the sensor signal (bandpass range:
It is in this sense that NPF control is useful particularly when
0.1 Hz and 100 kHz), the controller −H (jω) was implemented
multiple modes are to be controlled.
digitally in a target PC with a sampling frequency of Fs =
32 kHz using the Matlab/Simulink xPC Target module [4],
3. Experimental work an anti-aliasing low pass filter (Krohn-Hite 3384) was used
with a cutoff frequency of 16 kHz, analogue circuits (the
3.1. Plant identification
voltage follower and the phase inverter) were manufactured
A smart beam similar to that shown in figure 1 was constructed to supply the PZT actuators with signals of opposite phase
using a cantilever beam of aluminum (330 mm × 61 mm × to generate a pure bending moment in the beam, and finally
3 mm) embedded with a matched pair of PZT ceramic patch two voltage amplifiers K a (PCB 790A01) were used to excite
actuators (Fuji Ceramics Co., C-83H, 60 mm × 25 mm × 1 mm) the PZT actuators. The detailed wire-connections to the PZT

5
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

Figure 3. Open-loop FRFs (a), reduction ratios (b), displacement (c) and velocity (d) responses of a two DOF structure of ςs2 = 0.01,
δ21 = 1, and α21 = 0.1. Refer to figure 2 for the detailed captions.

Figure 4. Cantilever beam (330 mm × 61 mm × 3 mm) embedded


with a matched pair of PZT ceramic patch actuators
(60 mm × 25 mm × 1 mm) and a PVDF patch sensor
Figure 5. Feedback loop circuit and a cross-sectional view of the
(60 mm × 12 mm × 0.028 mm) realizing the schematic plot in
beam embedded with a pair of PZT actuators and a PVDF sensor.
figure 1, where the hammer was used to excite the beam and the
accelerometer was attached on the beam to monitor the end tip
vibration before and after control.
guarantee of stability was tested. All the measurement results
presented are for full volume.
actuators are also indicated in figure 5 together with the poling The plant FRF G(jω) was measured from the path
directions. A manually controllable potentiometer (volume between the volume controller and the sensor preamplifier
controller) was also used when a new controller with no in figure 5. Rather than the plant itself, the generalized

6
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

3.2. Controller design and tests


From the identified model in figure 6, it is straightforward
to design the NPF and PPF controllers from equations (13),
(14), (16) and (17). Once the filter parameters (ωa,n , bn , and
ka,n ) were calculated, the discrete forms of the controllers were
implemented in the xPC target using the filter forms given
in appendix A [4]. The fundamental mode was first treated.
Figures 7(a) and (b) show (a) the Nyquist plot of the open-
loop FRFs of the NPF (solid line) and PPF (dashed) controllers
together with the boundary circle of the 2 dB control spillover
(dash-dot) and (b) the end tip acceleration responses before
(dotted) and after control (solid and dashed), respectively. The
corresponding reduction ratios and the PVDF sensor responses
before and after control were also calculated and measured in a
similar way to figure 2, but are not shown here for brevity. The
MOCC of the NPF control was the second mode while that
of the PPF control occurred due to the static robustness. The
circular locus (solid line) of lMOC ≈ 0.46 in figure 7(a) was
due to the MOCC with the resonance of 134 Hz. The results
are similar to those shown for an ideal single DOF system in
figure 2; subtle differences in the fundamental modal responses
were due to mutual coupling with the higher order structural
modes. The PPF control appears to perform better, but this is at
the expense of increased control spillover at lower frequencies
below the first resonant region. If the higher modes are more
remotely located from the fundamental mode with a smaller
amplitude, the NPF controller will perform better. Thus, the
performances of NPF and PPF control are strongly dependent
on the plant under control. The second mode was then treated
and the results are shown in figures 7(c) and (d) in the same
Figure 6. Measured (solid line) and modeled (dashed) generalized form of figures 7(a) and (b). The MOCC for the NPF controller
mobility forms jωG(jω) of the plant G(jω): (a) amplitude and
was the seventh mode above 2 kHz, while that for the PPF
(b) Nyquist plot. The first six modal responses (dotted) are also
overlaid. controller was the first mode. Thus the NPF controller was
designed based on the seventh mode (its maximum loop gain
lMOC was slightly larger than that for the sixth mode shown),
although not shown in the figure due to a limited frequency
mobility (solid lines) jωG(jω) is shown in figures 6(a) and (b)
resolution because of the bandwidth used for impact tests. The
together with its identified model (dashed) using six modes results are similar to those simulation results shown for the two
(dotted) which are also individually overlaid in figure 6(a). DOF system in figure 3. It is clear that the NPF controller
The agreement is good in both the amplitude and Nyquist performs better, but there was control spillover more than 2 dB
plots, and the Nyquist plot of the plant exhibits the minimum for the fifth and sixth modes. This result, which was not
phase characteristic (Re[jωG(jω)] > 0) up to the frequency predicted by the theory in section 2, was mainly due to the
range considered. Although not shown here, the mobility of time delays in the xPC target controller and the anti-aliasing
the plant above 2 kHz had two additional dominant peaks at low pass filter [4, 9] which could be greatly reduced if an
around 2600 Hz (seventh mode) and 3400 Hz (eighth mode) analogue control filter were used. The fundamental and second
and then had showed a noisy amplitude response loosing the modes were then controlled simultaneously and the results are
minimum phase characteristic. The relative modal amplitudes shown in figures 7(e) and (f). The control performance shown
are highly dependent on the length of the PZT actuators used. in figure 7(f) is similar to that achieved by superimposing
It is seen that, for the particular size of the PZT actuators figures 7(b) and (d). The control spillover of the sixth mode
used (2/11 of the beam length), the responses of the fifth to reaches up to 6 dB because of the time delay of the control
eighth modes turn out to be quite large. PPF control is known system as well as the multiple cross-coupling with the two
to be suitable for such a plant that does not roll-off at high absorber filters. Comparing figures 7(a)–(f), together with
frequencies. However, the NPF was also tested here since those simulations in figures 2 and 3, demonstrates that NPF
the bandpass filtering characteristic of this controller (roll-off control performs as well as PPF control and it is particularly
rate: 1/ω) could circumvent such stability issues at very high useful when multiple modes are controlled. Finally the NPF
frequencies. This is examined in section 3.2 with real-time and PPF control performances simulated and tested for all three
control tests. cases are summarized in table 1, where reduction ratios RR(min)

7
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

Figure 7. NPF (solid lines) and PPF control (dashed) of the fundamental mode ((a) and (b)), the second mode ((c) and (d)), and both the
fundamental and second modes ((e) and (f)): ((a), (c), (e)) Nyquist plot together with the control spillover boundaries of 2 dB (dash-dot), and
((b), (d), (f)) the accelerometer responses before (dotted) and after (solid and dashed) control.

were obtained from the acceleration responses before and after simplicity, when a single filter is used to control a single mode.
control. From equations (5) and (9b ), the reduction ratio can be written
The experimental beam considered in this paper was as
lightly damped, with the damping ratios of the first two modes RR (dB) = −20 log10 |1 + (ca /cs )|,
being around 1%. It is interesting to investigate how such
active modal control strategies would perform if the damping where cs is the damping of the structural mode and ca is that of
were increased by a factor of 10, say. Consider the case, for the absorber which can also be regarded as the filter gain. To

8
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

Table 1. NPF and PPF control performances simulated


( gnn = ca,n /cs,n ) and tested (reduction ratios RR(min) ).
Controller NPF PPF
Modal order(s) 1 2 [1, 2] 1 2 [1, 2]
gnn 6.3 6.9 [6.3, 6.9] 14.4 2.4 [14.4, 2.4]
RR(min) (dB) −15 −17 [−15, −17] −21 −10 [−21, −10]

Figure A.1. NPF control of a single degree of freedom oscillator and


achieve a 20 dB reduction, for example, the filter gain must be its analogous mechanical model.
ca = 9cs . If cs is small, ca does not have to be large. If cs is not
small, ca has to be large. Applying a large gain inevitably risks
the stability of the control system. The non-minimum phase
characteristics of the plant and the uncertainty at very high Table A.1. Three different ways to realize the absorber in
figure A.1(b): PAF (position–acceleration feedback), VVF
frequencies, which are negligible when the gain is small, can
(velocity–velocity feedback), and APF (acceleration–position
become very important. Here, note that, unlike low frequency feedback). (Note: S = ωs2 − ω2 + j2ςs ωs ω, A =
responses, very high frequency (asymptotic) responses of a ωa2 − ω2 + j2ςa ωa ω, 1/m s = ωs2 /ks = 2ςs ωs /cs , ka = m a ωa2 =
flexible structure may only be marginally (little) affected by 2ςa ωa ca , σ = m a /m
√ s , κ = ka /ks√
, g = ca /cs , ωs2 = ks /m s , ωa2 =
damping. Thus, increasing the mechanical damping may ka /m a , 2ςs = cs / m s ks , 2ςa = m a ka /ca,n .)
not automatically suppress the threat to stability at very high Structure G(jω) Absorber H (jω)
frequencies. To achieve a large reduction with a moderately
damped plant, an additional shaping filter (i.e., a low pass filter) PAF (1/m s )S−1 ka (−ω2 )A−1
VVF (jω/m s )S−1 ka (jω)A−1
can be employed to minimize the threat to stability at very high
APF (−ω2 /m s )S−1 ka A−1
frequencies. However, it should be noted that this can cause
another threat to stability because of the transition region of
the shaping filter. Regarding robustness to a potential stability
threat at very high frequencies, PPF control (roll-off rate: ω−2 ) analogous mechanical model (consisting of m a , ca , and ka )
has an advantage over NPF control (roll-off rate: ω−1 ). shown in figure A.1(b) [12]. The lower mass system is an
m  c  k system while the upper mass system is an m –c–
4. Conclusions k system, where  denotes a parallel connection and − denotes
a serial connection, the damping ratios in equations (4a )
Control of the first two modes of a smart beam embedded with and (4b ) are correspondingly defined. The measured response
a matched pair of PZT actuators and a collocated PVDF sensor for the feedback controller can be displacement, velocity, or
using two modal control methods, NPF and PPF control, has acceleration. The corresponding controller forms that yield
been compared. It has been shown that NPF control is an the same open-loop gain in equation (7) as well as the same
electrical realization of a dynamic vibration absorber. The analogous model in figure A.1(b) are given in table A.1
focus has been on the robustness of the two controllers and the with the names: position–acceleration feedback (PAF) for
optimal parameters have been determined analytically, to give position feedback, velocity–velocity feedback (VVF) for
the optimal performance for NPF control and a near-optimal
velocity feedback, and acceleration–position feedback (APF)
performance for PPF control. The experiments conducted with
for acceleration feedback, where the PAF corresponds to NPF
a cantilever beam embedded with a matched pair of PZT patch
actuators and a collocated PVDF patch sensor demonstrate in this paper. Note that the equivalence, ka = m a ωa2 =
that it is possible to realize an electrical dynamic absorber 2ςa ωa ca , enables convenient implementation of the filter in
using the generalized displacement sensor. It has been further terms of any of the parameters (m a , ca , and ka ). Table A.2
demonstrated that, with the particular plant tested, NPF has gives the continuous and discrete time responses of the three
particular advantages in terms of performance and stability filters in table A.1, where the discrete forms [12] are obtained
when multiple modes of a structure are controlled. by applying the impulse invariant method. The discrete form is
used for the xPC target implementation discussed in the main
text. Once the filter gain h is known from analysis in which
Appendix A. Equivalence of various realizations of
the electrical dynamic absorber h = ka , m a or ca , the digital gain in the xPC target controller is
determined by
Figure A.1(a) illustrates the case when a single NPF filter
K d = h(K T )−1 , (A.1)
is used to control a single structural mode or a single DOF
oscillator (consisting of m s , cs , and ks ), where f p is the
external primary force, fc is the control force, and e is the where K = K a K s with the actuator and sensor gains K a and
structural response after control. If the control filter is of K s , and T = Tact Tsen with the actuator and sensor sensitivities
the form in equation (2), the filter can be represented by the Tact and Tsen , respectively.

9
Smart Mater. Struct. 20 (2011) 015011 S-M Kim et al

Table A.2. Continuous and discrete time responses of the absorbers in table A.1. (Note: ωd = ωa 1 − ζa2 , A(t) = exp(−ζa ωa t),

ϕ = tan−1 ( 1 − ςa2 /ςa ),  = z 2 − z 2e−ςωn T cos ωd T + A 2 , A = exp(−ζa ωa T ), and the sampling time is T = 1/Fs with Fs being the
sampling frequency.)
h(t) H (z)
1 1 z A sin ωd T
Pos. ka A(t) sin ωd t, ka
ωd ωd 

ωn ωn z(−z 1 − ςa2 + A sin(ωd T + ϕ))
Vol. −ka A(t) sin(ωd t − ϕ) −ka
ωd ω 
   d  
ω2 ωn2 z(−z 2ςa 1 − ςa2 + A sin(ωd T + 2ϕ))
Acc. ka δ(t) + n A(t) sin(ωd t − 2ϕ) ka 1 +
ωd ωd 


Appendix B. Maximally flat frequency response b= 2σ (1 − m)m −2 , (B.5)
strategy
where m = 1 √
2
− r . Equations (B.3) and (B.5) are equated
2

This appendix is a summary of the part of [11] that is applicable when r 2 = ( 5 − 1)/2, which corresponds to a control
to this paper. Consider the two DOF vibration system shown in spillover of about 2 dB.
figure A.1(b) and assume that ωa = ωs . The coupled structural
mobility (the velocity e of the lower mass system to force fp , References
after attachment of the upper mass system) can be written as
[1] Fanson J L and Caughey T K 1990 Positive position feedback
e 1 a(b + jx) control for large space structures AIAA 28 717–24
YCS (jx) = = , (B.1)
fp cs (a + jx)(b + jx) + σ [2] Balas M J 1979 Direct velocity feedback control of large space
structures J. Guid. Control 2 252–3
where σ = m a /m s , cs , a and b are defined in the main text [3] Friswell M I and Inman D J 1999 The relationship between
positive position feedback and output feedback controllers
and x =  − −1 is a normalized frequency bandwidth.
Smart Mater. Struct. 8 285–91
Den Hartog’s fixed-point theory [6] with the aim of achieving [4] Kim S M, Pietrzko S and Brennan M J 2008 Active vibration
the maximally flat coupled mobility FRF response can be isolation using an electrical damper or an electrical dynamic
described mathematically by absorber IEEE Trans. Control Syst. Technol. 16 245–54
[5] Moheimani S O R, Vautier B J G and Bhikkaji B 2006
|YCS (0)| = |YCS (jx c )|, (B.2) Experimental implementation of extended multivariable PPF
control on an active structure IEEE Trans. Control Syst.
Technol. 14 443–55
where the normalized
√ bandwidth between the two fixed points [6] Den Hartog J P 1947 Mechanical Vibration (New York:
is given by x c = ν + (σ/2) in which ν = ab . Combining McGraw-Hill)
equations (B.1) and (B.2) gives [7] Lee Y S, Gardonio P and Elliott S J 2002 A matched
√ piezoelectric sensor and actuator pair for vibration control of
b = 2σ . (B.3) a smart beam J. Acoust. Soc. Am. 111 2715–26
[8] Cunningham D R and Stuller J A 1995 Circuit Analysis
Since σ = κ with the assumption ωa = ωs , equation (14) 2nd edn (New York: Wiley) pp 441–59
[9] Kim S M, Elliott S J and Brennan M J 2001 Decentralized
in the main text is determined. The normalized optimal filter control for multichannel active vibration isolation IEEE
gain g = ca /cs for a given bandwidth b , which achieves the Trans. Control Syst. Technol. 9 93–100
best performance without violating the robustness constraint [10] Kim S M and Brennan M J 1999 A compact matrix formulation
of degree r , is given by using the impedance and mobility approach for the analysis
of structural-acoustic systems J. Sound Vib. 223 97–113
 [11] Kim S M, Wang S and Brennan M J 2011 Dynamic analysis
g = ((η + 1)(1 − r 2 )/r 2 )(1 + 1 − r 2 ((η − 1)/(η + 1))),
and optimal design of a passive and an active piezo-electrical
(B.4) dynamic vibration absorber J. Sound. Vib. 330 603–14
where η = μ/ν in which μ = (a 2 + b 2 )/2 and ν = ab . [12] Kim S M 1998 Active control of sound in structural-acoustic
Rearranging equation (B.4) with assuming a b gives coupled systems PhD Thesis Univ. of Southampton

10

Você também pode gostar