You are on page 1of 16

Auxiliary Command Input Shaping Technique to

Reduce Disturbance Induced Vibration


N. Medlinger, N. Sela and P-O Gutman
Faculty of Civil and Environmental Engineering
Technion - Israel Institute of Technology
March 28, 2017

Abstract
Input shaping is a method in which a reference command to a dynamic system
is shaped via an open-loop shaper, aiming to reduce the residual vibration of the
commanded system. In this work we seek to reduce disturbance-incurred residual vi-
brations, rather then input-induced vibrations. Input shaping techniques are utilized
to introduce an auxiliary command to the system once the disturbance is applied.
The time constant of the suggested shaper is chosen arbitrarily, thus the response
time can be significantly shorter compared to similar shapers. We also show that the
shaper is optimal in terms of residual vibration. The performance of the shaper and
its sensitivity to unknown model and disturbance parameters are verified via tests
on a flexible system test-bed.

1 Introduction
One of the major challenges in system and control engineering is the vibration reduction
of flexible systems during control actions. In many applications, precise position control
and rapid motion is required. Residual vibration attenuation is essential for Accurate po-
sitioning. The most common method to reduce residual vibration is to use a feedback
control loop. Another method, which has received less focus, is the idea of shaping the
input in such a way that the vibration modes are not excited. Flexible structures would be
easier to control when vibrational modes are attenuated [1]. Thus, there is motivation to
pursue schemes which combine a pre-filter, to shape the actuator inputs, with a subsequent
open-loop or closed-loop scheme.
A common way used to apply a pre-filter to a controlled system is the input shaping [2]
method, in which a real time convolver convolves the reference input with a sequence of
impulses having appropriate magnitudes and time spacing. One of the drawbacks of tra-
ditional input shaping is due to its inability to suppress vibrations caused by external
disturbances. When using input shaping, vibration is effectively suppressed if of the vi-
bration is command-incurred, whereas vibrations caused by disturbances are hardly dealt
with. In this work we focus on utilizing the input shaping approach, aiming to decay
disturbance-incurred residual vibrations. Special attention will be given to a certain kind
of disturbance, namely, an impulse disturbance.

1
The term Input Shaping c was first introduced by Singer and Seering [2] and it refers
to the operation of convolving a desired input command with a set of impulse sequence.
Once the shaped command is introduced into a flexible dynamic system, the residual
vibration of the system is attenuated after the last impulse is applied. The idea is based
on the fact that applying an impulse, A1, to a flexible system will cause it to vibrate, but
if a second impulse, A2, is applied at a later time, then the vibration induced by the first
impulse may be canceled. The second impulse must be applied at the correct time and
must have the appropriate magnitude for complete cancelation.

Figure 1: Superposition of two impulse responses which yields cancelation and zero residual
vibration after the second impulse.

The amplitudes and time instances of the impulses in an input shaper are determined
by solving a set of constraint equations. A variety of constraints are used [1] and [3],
among them are robustness constraints [2], residual vibration constraints [4], [5], impulse
amplitude constraints, [6] and the requirement of time optimality [7]. In [2] a simple shaper
with only residual vibration and impulse amplitude constraints, called ZV-shaper is shown.
The ZV-shaper possess two impulses with the following properties
 
1 K
Aj = , , tj = [0, 0.5Td ] , j = 1, 2 (1)
1+K 1+K
where n and are the system natural frequency and damping ratio respectively, K =
p
e 12 , Td = 2d is the systems damped period, and d = n 1 2 is the damped
natural frequency. Notice that the ZV-shaper time length, t2 , depends on the systems
damped period, and can not be shortened arbitrarily. This is a major weakness of these
kind of shapers, since they incur a costly time penalty. A limited attempt to overcome
this drawback is presented in [6], where negative impulses are allowed in the filter. This
alleviates the problem to some extent, since the shaper time instances are still depended
on systems period. A different approach is taken in [8] where the Optimal Arbitrary
Time-delay Filter (OATF) is introduced. The OATF is a 3-impulse filter given by
3
X
ISOAT F (t) = Aj (t (j 1) ) (2)
j=1

2
where is the filter time-delay constant which is chosen arbitrarily. By formulating an
optimal problem, with the residual vibration as the cost function and Aj as the decision
variables, the filter amplitudes are calculated implicitly. Similar to the shaper in [6], the
OATF incorporates a negative impulse.
All of the work mentioned above deals with command incurred residual vibration, much
less attention is given to reducing vibrations due to external disturbances. Aiming to sup-
press vibrations caused by external disturbances the CIST - Commandless Input Shaping
Technique, is introduced in [9] and [10]. According to CIST, once the disturbance is
detected, a set of impulses is directly applied to the system via an actuator in order to
suppress the induced vibration. The main weakness of the CIST concept is the fact that
dynamic systems can not be controlled by direct impulses, thus, the CIST seems to be
impractical. In [11] input shaping is applied to a dynamic system with non-zero initial
conditions. The problem is formulated as an optimal control problem and the solution is
shown to be a bang-bang. Although this may be seen as a system with impulse disturbance,
the solution in [11] is restricted to the bang-bang case.
In this work we will focus on applying input shaping approaches to flexible dynamic
systems aiming to decay disturbance-incurred residual vibrations. Special attention will
be given to a certain kind of disturbance, namely, an impulse disturbance. In section 2
the main ideas of the suggested input shaping technique, together with sensitivity analysis,
are drawn. Section 3 encloses validating test results, and concluding remarks are given in
section 4

2 The ACIST - Auxiliary Command Input Shaping


Technique
Following the footsteps of [9] we introduce a novel approach to reduce residual vibration
induced by impulse-like disturbances - Auxiliary Command Input Shaping Technique -
ACIST. The idea is to induce an auxiliary reference command rA (t) once the disturbance
is detected. The auxiliary command is convolved with a properly designed input shaper
to reduce the residual vibration. In Figure 2, a block diagram of the ACIST concept is
shown.

d(t)
ACIST

rA (t) ACISTCmd (t) y(t)


IS G(s)

Figure 2: ACIST concept.

Referring to Figure 2 and assuming linearity, the output, y(t), can be represented by
the following sum:
y(t) = yd (t) + yACIST (t) (3)

3
where yd (t) is the systems response to the disturbance, d(t), and yACIST (t) is the response
of the system to the shaped auxiliary reference command, ACISTCmd (t). We will focus on
an impulse-like disturbance, occurring at time td and with magnitude Ad , thus:
d(t) = Ad (t td ) (4)
Consider G(s) to be a linear harmonic oscillator with natural frequency n and damping
ratio , which obeys the dynamic state-space equations:

A
z
 }| {  
0 1 0
~x (t) = 2 ~x(t) + 2 u(t)
n 2n
| {zn }

B

y(t) = 1 0 x(t) (5)
| {z }

C

where ~x = [x1 x2 ]T is the state vector, x1 being the oscillators mass position and x2 = x 1
its velocity. Thus, the response yd (t) is [2]
Ad n n (ttd )
yd (t) = p e sin (d (t td )) (6)
1 2
We seek to bring the total response, y(t), to zero after some finite time t tn by a well
designed ACIST.

2.1 Simple ACIST


We will begin with the simplest form of the ACIST, in which the auxiliary command
rA (t), is a step command of magnitude RA at time tr and the shaper, IS(t), is a set of two
impulses with amplitudes A1 , and A2 , and times t1 and t2 respectively:
IS(t) = A1 (t t1 ) + A2 (t t2 ) (7)
Let yACIST (t) be the response of G(s) to the ACIST command, then, yACIST (t) is defined
by the inverse Laplace transform
 
1 RA s(tr +t1 ) s(tr +t2 )

yACIST (t) = L A1 e + A2 e G(s) (8)
s
Substituting (5) and (8) into (3), and setting td = tr = t1 = 0 yields the total response,
y(t)
Ad n en t
y(t) = RA (A1 + A2 ) + p sin (d t) +
1 2
" #
RA A1 en t sin (d t)
+ cos (d t) + p +
2 2 1 1 2
" #
RA A2 en (tt2 ) sin (d (t t2 ))
+ cos (d (t t2 )) + p (9)
2 2 1 1 2

4
In order to eliminate the residual vibration, we must choose RA , A1 , A2 , and t2 in a way
that would bring y(t) = 0. From (9) it is clear that we must set A1 = A2 . In addition,
we can see that RA can be chosen arbitrarily. thus, only A1 and t2 are unknown. Utilizing
a trigonometric identity we can rewrite (9)

y(t) = RA en t C 2 + S 2 sin (d t + ) (10)
where
Ad n A1 " #
RA 1 2A1 en t2   cos( t )
d 2
C=p + cos d t2 + p
1 2 2 2 1 2 2 1 2 1 2
" #
A1 A1 en t2   sin( t )
d 2
S= 2 2
sin d t2 + p (11)
2 1 2 1 2 1 2
For zero vibration we set C = 0 and S = 0, which yields the following equation set
" #
Ad n (2 2 1) cos( d t2 )
p +p = en t2 sin (d t2 ) + p (12)
RA A1 1 2 1 2 1 2
" #
n t2 sin(d t2 )
1=e cos(d t2 ) p (13)
1 2
First we solve (13) numerically to get t2 , then, by substituting it into (12), we get an
explicit solution for A1
Ad n (3 2 2 4 1)en t2
A1 = p (14)
RA 1 2 sin(d t2 )
Simulation results of the simple ACIST vibration reduction, compared to the no shaping
case, are shown in Figure 3. The shaper consists a set of two impulses, and the system
parameters are: = 0.01, n = 2. The second impulse time, t2 , was found to be:
t2 = 0.944[sec]. The input disturbance impulse is imposed by a velocity initial condition
of x2 (0) = n2 which is equivalent to an amplitude of Ad = 1.
Notice that for zero damping, the set (12)-(13) reduces to
Ad n
= sin (d t2 ) (15)
RA A1
1 = cos(d t2 ) (16)
n
for which the solution is t2 = d
, where n = 0, 2, 4, ..., and A1 .

From Figure 3 it is clear that the suggested ACIST, in its simplest form, reduces
the disturbance induced residual vibration drastically for t > t2 . However, this reduction
comes with a heavy cost, since for the zero damping case the ACIST is un feasible, and
for the damped case, the magnitude of the transient response is much higher than the re-
sponse without ACIST. The reason for this undesired transient lies in the structure of the
ACIST impulse set. Since the ACIST includes only two impulses, the second impulse,
which counteracts the motion of the dynamic system, G(s), due to the shaped input and
the disturbance, can only be applied at time t2 . By then, the total transient response
reaches a significant level. Since t2 depends only on d and , it can not be arbitrarily
shortened.

5
ACIST Simulation Results
40
No ACIST
With ACIST
20

y(t)
0

20
0 0.5 1 1.5 2 2.5 3 3.5 4
t

20

15
(t)
SH

10
u

0
0 0.5 1 1.5 2 2.5 3 3.5 4
t

Figure 3: Simulation results of ACIST with a two impulse shaper and damping. Upper
graph: Response y. Lower graph: Shaped command, ACISTCmd . = 0.01, n = 2, the
disturbance is modeled by velocity initial condition of x2 (0) = n2 .

2.2 ACIST with OATF


Aiming at implementing the second impulse earlier, in order to reduce the transient re-
sponse, we suggest to enhance the ACIST by using a set of three impulses based on the
Optimal Arbitrary Time-delay Filter [8]
3
X
ISOAT F (t) = Aj (t (j 1)) (17)
j=1

were is the filter time-delay constant which is chosen arbitrarily. Substituting (17)
into (8) gives

RA  
yACIST (t) = L1 A1 es(tr ) + A2 es(tr +) + A3 es(tr +2) (18)
s
Recall that td = tr = 0. The total response of the dynamic system, G(s), to the ACIST
command and the disturbance, d(t), is

y(t) = RA (A1 + A2 + A3 ) + RA en t C 2 + S 2 sin (d t + ) (19)

6
where
A1 " #
Ad n 1 2 A2 en cos(d )
C= p + + 2 sin(d ) + p +
RA 1 2 2 2 1 2 1 1 2
" #
A3 e2n cos(2d )
+ sin(2d ) + p
2 2 1 1 2
" #
A1 A2 en sin(d )
S= 2 + 2 cos(d ) p +
2 1 2 1 1 2
" #
A3 e2n sin(2d )
+ cos(2d ) p (20)
2 2 1 1 2
P
For zero vibration we set C = S = 0. In addition, we must add the constraint j Aj = 0 to
prevent the ACIST from producing additional command after the last impulse is applied.
We sum-up to get the final equation-set:
0 = A1 + A2 + A3
" #
Ad n (2 2 1) A1
p =p + A2 en sin(d ) + p cos(d ) +
RA 1 2 1 2 1 2
" #

+ A3 e2n sin(2d ) + p cos(2d )
1 2
" #

0 = A1 + A2 en cos(d ) p sin(d ) +
1 2
" #

+ A3 e2n cos(2d ) p sin(2d ) (21)
1 2
The solution of (21) for the undamped case is
A1 = K (1 + 2 cos(n )) (22)
A2 = 2K (1 + cos(n )) (23)
A3 = K (24)
Ad n
where K = 2RA sin(n )
, and for 6= 0 is

en cos(d ) + ` sin(d )
A3 = KK (25)
2 cos(d ) (en + en )
 
1e 2n `
cos(2d) sin(2d )
A2 = A3   (26)
1e n `
cos(d ) sin(d )
A1 = A2 A3 (27)

and ` =
2 2 1 Td
where K = . Notice that for = 2
the term in the denominator
1 2 1 2
of K tends to zero, which means that A3 . Simulation results of the ACIST with
OATF input shaper vibration reduction are shown in Figure 4.

7
ACIST with OATF Simulation Results With Damping
10
No ACIST
5 With ACIST

y
0

5
0 0.5 1 1.5 2

5
ACISTCmd

10
0 0.5 1 1.5 2
Time [sec]

Figure 4: Simulation results of ACIST with OATF shaper with damping. Response y, -
upper. Shaped command, ACISTCmd , - lower. = 0.1, n = 2[rad/sec] and = T2d [sec].
Disturbance is modeled by the initial condition x20 = n2 .

2.3 ACIST via Optimization


The results of the previous section can also be obtained by the following minimization
procedure. The solution x(t) of (5) can be written by:
Z t 3
X
0)~x(0) + RA
~x(t) = (t,
(t, )B Aj (t (j 1))d (28)
0 j=1

) is the transition matrix from to t


where (t,

) = eA(t )
(t, (29)
solving the integral in (28) for t 2 gives
0)~x(0) + RA A1
~x(t) = (t, [0,] (t) + RA (A1 + A2 )
[,2] (t) (30)
where
Z tf
[t ,t ] (t) =
)Bd
(t,
f i
ti
 
`
n (ttf ) sin(d (t tf )) + cos(d (t tf ))
=e d
1 2 sin(d (t tf ))
 
`
n (tti ) sin(d (t ti )) + cos(d (t ti ))
e d (31)
1 2 sin(d (t ti ))

8
Substituting (31) into (30) yields
0)~x(0) + (t)
~x(t) = (t, f~A (32)
 

where (t) = [2] (t)
[0] (t),
[2] (t) [] (t) , f~A = [A1 , A2 ]T is the vector of the
ACIST magnitudes (including RA ), and [2] (t) is
 
`
n (ttk ) sin(d (t tk )) + cos(d (t tk ))

[tk ] (t) = e d (33)
1 2 sin(d (t tk ))

To get zero residual vibration at t = 2 we wish to set ~x(2) = [0, 0]T . We minimize the
following cost function
1 ~x(2)
J = ~x(2)T W (34)
2
with W some weighting matrix. Substituting (32) into (34) and solving for t = 2
1 T

(2,
J= (2, 0)~x(0) W 0)~x(0)+
2
1  T T
+ (2)f~A W (2)
f~A + (2,
0)~x(0) W (2)
f~A (35)
2
Differentiating (35) with respect to f~A and equating to zero will give
 1 h T i
f~A = (2)
T
W (2)
(2, 0)~x(0) W (2)
(36)

A direct solution for (36) with W = I will give the same results as in (22) and (25).
2
Furthermore, it can be shown that the second optimality condition ~J2 is fulfilled when
fA

sin(n ) 6= 0, the same holds for (2) T
W (2) to be invertible.

2.4 ACIST Sensitivity Analysis


In order to investigate the sensitivity of the ACIST to system and disturbance parameters,
we expand (18) to capture the response to a set of N impulses, which is now:
N
X
y(t) = RA Aj + RA en t C 2 + S 2 sin (d t + ) (37)
j=1

where N is the number of impulses in the sequence and


N
" #
Ad n en td cos(d td ) X Aj en tj cos(d tj )
C= p + sin(d tj ) + p
RA 1 2 j=1
2 2 1 1 2
N
" #
Ad n en td sin(d td ) X Aj en tj sin(d tj )
S= p + 21
cos(d tj ) p (38)
RA 1 2 j=1
2 1 2

To quantify the residual vibration level of a system we use the vibration error Verr as in [4]
and [5]
RA C 2 + S 2
Verr = (39)
Ad n 2
1

9
which is the amplitude of the residual vibration of the system subject to the impulse dis-
turbance with ACIST, divided by the amplitude of the residual vibration of the response
to impulse disturbance only. It would be useful to investigate the robustness of the ACIST
to the system parameters, n and , and to the disturbance parameters, Ad and td .
First we will verify to what extent is the ACIST sensitive to uncertainty in the distur-
bance amplitude, Ad . Assuming = 0 and td = 0 (38) reduces to
N
Ad n X
C= Aj sin(d tj )
RA j=1
N
X
S= Aj cos(d tj ) (40)
j=1

Since the ACIST parameters, Aj and tj , are calculated such that C and S in (40) will
equal zero, we can state that
N
Ad n X
= Aj sin(d tj )
RA j=1
N
X
0= Aj cos(d tj ) (41)
j=1

where Ad is the estimated disturbance amplitude. Substituting (32) into (39) yields

A A
d d
Verr = (42)
Ad
A sensitivity curve for the Ad estimation error is plotted in Section 3.

Second, we will look into ACIST sensitivity to uncertainty in td . When td is exactly


known, we set td = t1 = 0, and the ACIST is initiated at the same time the the disturbance
is introduced. By setting td 6= 0 and leaving t1 to equal zero, we are actually implementing
the ACIST not at the correct time, td , but at a wrong time, t1 , which means uncertainty
in td . Assuming = 0 but leaving td 6= 0 (38) reduces to
N
Ad n X
C= cos(d td ) Aj sin(d tj )
RA j=1
N
Ad n X
S= sin(d td ) Aj cos(d tj ) (43)
RA j=1

Since the ACIST parameters, Aj and tj , are calculated such that C and S in (38) will
equal zero, we get
N
Ad n X
= Aj sin(d tj )
RA j=1
N
X
0= Aj cos(d tj ) (44)
j=1

10
Substituting (44) into (43) yields

Ad n
C= (cos(d td ) 1)
RA
Ad n
S= sin(d td ) (45)
RA
Thus, we get the following vibration error

rA C 2 + S 2 p
Verr = = 2 (1 cos(d td )) (46)
Ad n
A sensitivity curve for the td error is plotted in Section 3.

Third we explore the ACIST sensitivity to uncertainty in the system natural frequency,
n , which is more cumbersome since there is no closed form solution for any ACIST filter.
However, it is quite straightforward to calculate this sensitivity numerically for a given filter.
For example, for the ACIST with OATF the sensitivity to n uncertainty is obtained as
follows: we set = 0, since its influence on the sensitivity to n uncertainty is negligible,
and solve (21) by setting C = S = 0 to get A1 , A2 and A3 , but instead of the true natural
frequency, n , we use the estimated natural frequency, n , which yields
Ad
n
A1 = A3
RA tan( n )
A2 = A1 A3
 

n Ad 1 1
A3 = (47)
2RA (cos(n ) 1) sin(
n ) tan(
n )

By substituting A1 -A3 we just found into (38) we calculate back C and S using the true
natural frequency, n . Finally, we substitute the calculated C and S into (39) together
with the true natural frequency, n , to get the sensitivity,Verr .

The sensitivity analysis for modeling error of the damping ratio, , is similar to the
natural frequency sensitivity analysis, thus, it is omitted here.

3 Results
To validate the results we used a simple test-bed shown in Figure 5. The test-bed includes
a flexible rod connected to a dc-motor, encoder and a solenoid. The dc-motor, encoder and
solenoid are connected to a PC on which a position controller is implemented. The rod
is preloaded by the solenoid. A command from the PC extracts the solenoid and the rod
starts to rotate in closed-loop. When the rod passes through the zero-line for the first time,
the ACIST command is generated and sent to dc-motor from the PC. The time needed for
the rod to reach the zero-line after solenoid extraction is measured off-line. To measure the
residual vibration of the rods tip, an accelerometer is attached to the tip and a fast rate
camera is stationed above it. The ACIST parameters are calculated from an identified, 2nd
order transfer function, of the test-bed closed-loop. In Figure 6 test results of the ACIST

11
with three impulse OATF input shaper disturbance induced vibration reduction, compared
to a test without the ACIST shaper, are shown. The closed-loop system parameters are:
= 0.07, n = 2.3[rad/sec], and was chosen to be: = Td [sec].
In Figures 7, 8 and 9 the ACIST sensitivity to Ad estimation error, sensitivity to
error in td , and sensitivity to n modeling error, calculated from the analytic solution
and measured with the test bed, are plotted. Notice that td error is normalized by the
system damped period, Td . Also notice that we had to add an 8% measurement error to
the analytic sensitivity curve. This is because the analytic sensitivity calculation assumes
zero residual vibration when every thing is perfectly known. Yet, zero residual vibration is
practically impossible due to affects like measurement error and higher modes dynamics.
A good correlation between the sensitivity analysis and the sensitivity tests is apparent
from Figures 7- 9. An exception to that is the sensitivity to large values of td error. The
considerable difference between the analytic solution and the test results for large values
of td error may be explained by actuator saturation, which prevents the generation of high
acceleration (current) and high velocity. These limiting effects are not taken into a count
in the analytic analysis.

zero-line

dc-Motor+Encoder

Rod

Solenoid

Accelerometer

Figure 5: Test-bed.

12
ACIST with OATF Test Results
200

Tip Pos [Pixel]


No ACIST
With ACIST
0

200
0 0.5 1 1.5

10
ACISTCmd [V]

0 0.5 1 1.5
Time [s]

Figure 6: Test results of ACIST with OATF shaper. Tip position-upper. ACISTCmd -
lower. Using closed-loop parameters: = 0.07, n = 2.3[rad/sec] and = Td [sec].

ACIST with OATF sensitivity to Ad Error


80

70

60 Analytic
Analytic+Msr Err
50 Measured
Verr %

40

30

20
8% Msr Err
10

0
0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Normalized Distubance Inpulse Amplitude Ad /Ad

Figure 7: Calculated and measured sensitivity of ACIST with OATF shaper to Ad error.
Using closed-loop parameters: = 0.07, n = 2.3[rad/sec] and = Td [sec].

13
ACIST with OATF Sensitivity to td Error
200

150
Analytic
Analytic+Msr Err
Measured
Verr %
100

50

8% Msr Err
0
0.1 0 0.1 0.2 0.3 0.4
Normalized Disturbance Impulse Time Error td /T

Figure 8: Calculated and measured sensitivity of ACIST with OATF shaper to error in
td error. Using closed-loop parameters: = 0.07, n = 2.3[rad/sec] and = Td [sec].

ACIST with OATF Sensitivity to n Modeling Error


100

Analytic
80 Analytic+Msr Err
Measured
60
Verr %

40

20
8% Msr Err

0
0.6 0.7 0.8 0.9 1 1.1 1.2
Normalized Frequency
n/n

Figure 9: Calculated and measured sensitivity of ACIST with OATF shaper to n mod-
eling error. Using closed-loop parameters: = 0.07, n = 2.3[rad/sec] and = Td [sec].

14
4 Conclusions
A technique to reduce the residual vibration of a flexible system was presented. The
technique is based on the input shaping concept, and it is a feed-forward approach that
can enhance closed-loop feedbacks. Unlike previous work, which utilize input shaping to
reduce vibration caused by reference commands, our method deals with residual vibrations
caused by an external, impulse-like, disturbance. The shaper parameters are obtained via
a simple algebraic solution based on the disturbance and system parameters. It was shown
that the shaper is optimal in terms of residual vibration. Analytic analysis and hardware
experiments verified the technique performance and its robustness to both, system and
disturbance, uncertainties.

References
[1] William Singhose. Command shaping for flexible systems: A review of the first 50
years. International Journal of Precision Engineering and Manufacturing, 10(4):153
168, 2009.
[2] Neil C Singer and Warren P Seering. Preshaping command inputs to reduce system
vibration. Journal of Dynamic Systems, Measurement, and Control, 112(1):7682,
1990.
[3] Tarunraj Singh and William Singhose. Input shaping/time delay control of maneu-
vering flexible structures. In American Control Conference, 2002. Proceedings of the
2002, volume 3, pages 17171731. IEEE, 2002.
[4] William E Singhose, Lisa J Porter, Timothy D Tuttle, and Neil C Singer. Vibration
reduction using multi-hump input shapers. Journal of dynamic systems, Measurement,
and control, 119(2):320326, 1997.
[5] William Singhose, Steve Derezinski, and Neil Singer. Extra-insensitive input shapers
for controlling flexible spacecraft. Journal of Guidance, Control, and Dynamics,
19(2):385391, 1996.
[6] W Singhose, Neil C Singer, and Warren P Seering. Design and implementation of time-
optimal negative input shapers. In Proceedings of the 1994 International Mechanical
Engineering Congress and Exposition, pages 151157. Citeseer, 1994.
[7] Lucy Y Pao and William E Singhose. On the equivalence of minimum time input shap-
ing with traditional time-optimal control. In Control Applications, 1995., Proceedings
of the 4th IEEE Conference on, pages 11201125. IEEE, 1995.
[8] David P Magee and Wayne J Book. Optimal filtering to minimize the elastic behavior
in serial link manipulators. In American Control Conference, 1998. Proceedings of the
1998, volume 5, pages 26372642. IEEE, 1998.
[9] Pyung-Hun Chang, Juyi Park, and Joon-Young Park. Commandless input shaping
technique. In American Control Conference, 2001. Proceedings of the 2001, volume 1,
pages 299304. IEEE, 2001.

15
[10] Joon-Young Park and Pyung-Hun Chang. Vibration control of a telescopic handler us-
ing time delay control and commandless input shaping technique. Control Engineering
Practice, 12(6):769780, 2004.

[11] Abhishek Dhanda, Joshua Vaughan, and William Singhose. Optimal input shaping
filters for non-zero initial states. In American Control Conference, 2009. ACC09.,
pages 41414146. IEEE, 2009.

16

You might also like