IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO.

9, SEPTEMBER 2013 3857

Robust Zero Placement for Motion Control of

Lightly Damped Systems
Chang-Wan Ha, Student Member, IEEE, Keun-Ho Rew, Member, IEEE, and Kyung-Soo Kim, Member, IEEE

Abstract—A unified tuning rule for the motion profiles is newly In particular, an optimized S-curve was proposed based on a
proposed to reduce the residual vibration. For several motion pro- frequency analysis of the forcing function input [13], [14]. In
files, including trapezoidal velocity profile, S-curve, and asymmet- addition, a nonlinear filter satisfying the velocity and acceler-
ric S-curve, the characteristics and tuning rules are analytically
formulated using the time- and Laplace-domain approaches. By ation constraints was proposed for smooth motion [15]–[17].
adjusting the tuning parameters to place the zeros of the motion Recently, the iterative learning control theory was adopted for
profile on a vibratory pole of the system, the residual vibration generating a position trajectory with minimized jerk [6], [18].
can be minimized. In particular, the placement of multiple zeros To enhance the performance of S-curve, other specially
significantly improves the robust performance against modeling designed motion profiles were suggested [19]–[21]. One of
errors, owing to robust pole-zero cancellation. By experiments, the
efficacy of the proposed approach is validated. the most practical and promising approaches is introducing
asymmetricity between acceleration and deceleration periods
Index Terms—Asymmetric S-curve (AS-curve), asymmetric- for effectively adjusting motion profile [22]. Owing to the
ity, motion profiles, residual vibration, robust zero placement,
S-curve, smoothness, trapezoidal velocity profile. additional degree of freedom, it is possible to generate a
fast-starting and slow-arriving motion (or slow-starting and
I. I NTRODUCTION fast-arriving motion). A complete closed-form solution to
asymmetric S-curve (the so-called AS-curve) was derived and
R EDUCING residual vibration for fast and accurate move-
ment has been one of the major issues in motion control
engineering for the last several decades. Ranging from con-
applied to many applications [23]–[25]. It turned out that, by
selecting appropriate parameters of AS-curve, the settling time,
as well as the residual vibration, can be remarkably reduced
sumer electronic devices to manufacturing equipment, motion
control techniques are widely adopted for enhancing the perfor- [26], while the tuning was heuristically conducted by a trial-
and-error method. More recently, in [27], a simple but analytic
mance and the productivity. For example, hard disk drives rely
method has been proposed to tune the design parameters of the
on the motion control profile for fast operation [1]. In addition,
semiconductor manufacturing equipment requires pick-and- AS-curve, based on the pole-zero cancellation. After converting
the AS-curve to a Laplace-transformed signal, the zeros of it
place motions having high speed and accuracy for enhanced
productivity and quality. can be placed on the vibratory pole of the system in the absence
In view of the practical implementation, the trapezoidal of modeling error. Unfortunately, however, the approach in [27]
suffers from the lack of robustness against the modeling error,
velocity profile is one of the simplest motion profiles. However,
compared with S-curve, it may induce more residual vibration, in particular, such as the natural frequency variation.
which has an adverse effect on the settling time [2], [3]. To tackle the robustness issue, this paper aims at newly
presenting a robust tuning rule against modeling errors. The
S-curve has been widely applied to motion control field (see,
e.g., [4]–[8]) and implemented using many methods, such as principle of robust pole-zero cancellation is analytically illus-
digital finite-impulse-response filter [9], quintic-spline differ- trated, and the physical characteristics are discussed. Moreover,
it will be shown that the proposed tuning rule can be applied to
ential analyzer [10], buttered digital differential analyzer [11],
and optimal control approach [12]. The major challenge with it the S-curve and the trapezoidal velocity profile, as well as the
AS-curve.
is to reduce the residual vibration in high-speed operations. To
In Section II, motion profiles introduced in literature (see
tackle this, important progresses have been made in literature.
[23]–[26]) are revisited. Then, in Section III, the characteristics
of the residual vibration are discussed. In Sections IV–VI,
Manuscript received September 26, 2011; revised April 8, 2012; accepted
June 18, 2012. Date of publication July 6, 2012; date of current version May 2,
a unified tuning rule is derived and verified by experiments.
2013. This work was supported in part by the Ministry of Knowledge Economy Finally, the conclusion follows in Section VII.
and Korea Institute for Advancement of Technology under the Human Resource
Training Project for Strategic Technology, and by the Ministry of Education and
Science Technology under BK21 Program, Republic of Korea. II. R EVIEW OF THE M OTION P ROFILES : T RAPEZOIDAL
C.-W. Ha and K.-S. Kim are with the Department of Mechanical Engineer-
ing, Korea Advanced Institute of Science and Technology, Daejeon 305-701, V ELOCITY P ROFILE , S- CURVE , AND AS-C URVE
Korea (e-mail: [email protected]; [email protected]).
K.-H. Rew is with the School of Mechanical Engineering, Depart- Recently, AS-curve, which is a superset of trapezoidal veloc-
ment of Robotics Engineering, Hoseo University, Asan 336-795, Korea ity profile and S-curve, has been introduced in a closed-form
(e-mail: [email protected]). solution, and its characteristics have been widely investigated
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. (see [23]–[26]). AS-curve has two tuning parameters for
Digital Object Identifier 10.1109/TIE.2012.2206334 adjusting the motion profile: smoothness and asymmetricity.

0278-0046/$31.00 © 2012 IEEE

3858 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 9, SEPTEMBER 2013

Fig. 3. Modeling of a lightly damped system.

The other tuning parameter γ may introduce the asymmetric-

ity between acceleration and deceleration periods. When γ = 1,
AS-curve turns out to be the S-curve. However, when the
asymmetricity is not equal to one, the fast-starting and slow-
arriving (or the slow-starting and fast-arriving) movements can
be obtained. This provides the flexibility in adjusting the motion
profile, as shown in Fig. 2(b).
Based on [26], given a target distance δtarget , if it holds that
(1 + β)(1 + γ)τm Vmax
δtarget ≥ δl := (3)
2
the profile is classified as a long distance case. In this case, one
may have the motion parameters as follows:
⎧
⎨ tj = βτm
ta = (1 − β)τm (4)
⎩ (δ −δl )
tv = target
Vmax

where tj , ta , and tv are the periods of constant jerk, constant

acceleration, and constant velocity, respectively. Interestingly,
Fig. 1. Notations of AS-curve [26]. there is a generic constraint between tj and ta such that

t j + t a = τm (5)

which implies that β plays a role of distributing the periods of

constant jerk and acceleration to achieve the maximum velocity
during the motion.
In this paper, only the long distance case will be discussed
for brevity. In addition, let us assume the following.
A1) Constant velocity period tv is long enough.
Fig. 2. Effects of the tuning parameters of AS-curve: (a) Smoothness and A2) The damping ratio ζ is negligible.
(b) asymmetricity [26]. Assumption A1) implies that only the vibration induced at
the deceleration period will be considered.
Notations of AS-curve and the effects of the two tuning param-
eters are shown in Figs. 1 and 2, respectively.
III. C HARACTERISTICS OF R ESIDUAL V IBRATION
The smoothness β ∈ (0, 1] determines the magnitude of the
applied jerk under the actuator limitations such as the maxi- The lightly damped system may be simplified as shown in
mum velocity Vmax and the maximum acceleration Amax . The Fig. 3. The dynamic relationship between the base and the
applied jerk is defined as second mass is described by the transfer function
Amax X2 −s2
J= (1) = 2 (6)
βτm X1 s + ωn2

where where X1 , X2 , and ωn are the absolute position of the base, the
relative position of the second mass with respect to the base,
Vmax
τm := . (2)  the natural frequency of the lightly damped system (ωn =
and
Amax k/m), respectively.
By controlling the base, suppose that x1 (t) exactly follows
As shown illustrated in Fig. 2(a), if the smoothness goes to zero the motion profile shown in Fig. 1. Then, from the acceleration
β → 0 (i.e., the jerk goes to infinity), the shape of the velocity profile for t ≥ t4 , it may be shown that
profile approaches the trapezoidal shape. On the contrary, as
the smoothness goes to one β → 1, the shape of velocity profile J (1 − e−βγτm s )(1 − e−γτm s )
s2 X1 (s) = − . (7)
approaches the bell shape. γ2 s2
HA et al.: ROBUST ZERO PLACEMENT FOR MOTION CONTROL OF LIGHTLY DAMPED SYSTEMS 3859

Fig. 4. (Top) Acceleration profile of the base movement and (bottom) the Fig. 5. Contour maps of the dimensionless residual vibration amplitude in the
induced vibration of the second mass during deceleration period. γ−β plane with fn τm = 1.587.

Hence, the relative motion of the second mass is given by the Hence, a dimensionless residual vibration amplitude, defined
following, from (6): as the ratio of the amplitude of the residual vibration to the
maximum acceleration, is expressed by the following, from
J (1 − e−βγτm s )(1 − e−γτm s )
s2 X2 (s) = . (8) Amax = Jβτm and ωn = 2πfn :
γ2 (s2 + ωn2 )
ẍ2 (t∗ )
Using the inverse Laplace transform, it is straightforward to a∗ :=
Amax
have
2
J = sin(πβγfn τm ) sin(πγfn τm ) . (12)
ẍ2 (t) = 2
[sin(ωn t) − sin (ωn (t − βγτm )) U (t − βγτm ) πβγ 2 fn τm
γ ωn
− sin (ωn (t − γτm )) U (t − γτm ) Fig. 5 shows the contour maps of the dimensionless residual
vibration amplitude drawn in the γ−β plane with fn τm =
+ sin (ωn (t − (1 + β)γτm )) 1.587 as example. The brightness of the color is proportional
× U (t − (1 + β)γτm )] (9) to the amplitude of the residual vibration (i.e., darker color
denotes less residual vibration). Note that there exist multiple
where U (t) is a unit step function. chasms where the residual vibration becomes much smaller
This shows that the relative acceleration of the second mass than the neighborhood. That is, adjusting the motion profile to
in the deceleration period can be expressed with four sine waves be in a chasm, the residual vibration can be minimized. The
with the same oscillating frequency but different phases, as conditions for chasm and its characteristics will be discussed in
shown in Fig. 4. The amplitude of the sine waves induced Section IV.
by an abrupt change of the applied jerk is determined by the
magnitude of the applied jerk and the natural frequency of the
lightly damped system. IV. F REE -V IBRATION C ONDITIONS
Our concern is to investigate the residual vibration of the The free-vibration conditions can be derived from (12) as
second mass after the motion. To this end, let us define a time reported in [27]. In particular, when either of the sine terms
shift as goes to zero, the residual vibration becomes zero. Hence, the
free-vibration conditions are as follows:
t∗ := t − (1 + β)γτm . (10)
C1. βγfn τ = k, k = 1, 2, 3, . . .
Then, when t∗ ≥ 0 (i.e., t ≥ t7 ), we have
C2. γfn τm = m, m = 1, 2, 3, . . . .
J
ẍ2 (t∗ ) = [sin (ωn t∗ + (1 + β)γωn τm )
γ 2 ωn
− sin(ωn t∗ + γωn τm ) A. Physical Meaning of the Free-Vibration Conditions
∗ ∗
− sin(ωn t + βγωn τm ) + sin(ωn t )] 1) C1: As shown in Fig. 6, when the time period between
   γω τ t4 and t5 (or t6 and t7 ), i.e., βγτm , is the integer multiples
4J βγωn τm n m
=− 2 sin sin of the period of oscillation (T = 1/fn ), the residual vibration
γ ωn 2 2
  is canceled out after the end of the base movement due to the
(1 + β)γωn τm destructive interference between four sine waves. The first sine
× sin ωn t∗ + . (11)
2 wave is canceled out by the second one. Similarly, the third sine
3860 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 9, SEPTEMBER 2013

Fig. 8. Pole-zero cancellations when (a) C1 and (b) C2 are satisfied with k =
m = 1, respectively.

poles of the relative vibration are

Fig. 6. (Top) Acceleration profile of the base and (bottom) response of the
second mass when free-vibration condition 1 is satisfied: βγτm = T . sp = ±jωn = ±j2πfn . (13)
To find the zeros in (8), one may have 1 − e−βγτm s = 0 at
2πk
sz1 = ±j , k = 0, 1, 2, . . . . (14)
βγτm
Also, the term 1 − e−γτm s gives the other set of zeros as
follows:
2πm
sz2 = ±j , m = 0, 1, 2, . . . . (15)
γτm
Therefore, in the case of AS-curve, there are one pair of
poles and two sets of zeros, determined by the characteristics
of the system and the designed motion control profile, respec-
tively. From (13) and (14), the nontrivial pole-zero cancellation
happens under C1. Also, from (13) and (15), C2 guarantees
the other pole-zero cancellation. Fig. 8 shows the pole-zero
cancellation when C1 and C2 are satisfied with k = m = 1. In
fact, the system poles are canceled by the zeros of the motion
profile, which effectively eliminates the vibratory mode of the
Fig. 7. (Top) Acceleration profile of the base and (bottom) response of the
second mass when free-vibration condition 2 is satisfied: γτm = T . system.

wave is canceled out by the fourth one. k implies the number C. Free-Vibration Conditions in the Special Cases
of oscillations within the time period between t4 and t5 (or t6
1) S-Curve: When γ = 1, S-curve is the special case of
and t7 ). The maximum vibration during deceleration period is
AS-curve [26]. The free-vibration conditions for S-curve are
fixed at J/(γ 2 ωn ), while residual vibration is eliminated after
obtained by substituting γ = 1 into C1 and C2. Then, we have
the end of the motion.
2) C2: As shown in Fig. 7, when the time period between βfn τm = k (16)
t4 and t6 (or t5 and t7 ), i.e., γτm , is the integer multiples of the
fn τm = m. (17)
period of oscillation, the residual vibration is canceled out after
the end of the motion due to the destructive interference. The Similar to the results of AS-curve, there are two free-vibration
first and second sine waves are canceled out by the third and conditions. However, contrary to the first free-vibration condi-
fourth ones, respectively. m implies the number of oscillations tion (16), the second one (17) may not be met with ease since
within the time period between t4 and t6 (or t5 and t7 ). fn and τm are not the free design parameters.
The maximum vibration during deceleration period is bounded 2) Trapezoidal Velocity Profile: When the acceleration and
within 2J/(γ 2 ωn ), while the residual vibration is eliminated deceleration periods are symmetric and the magnitude of the
after the end of the motion. applied jerk is infinite, the motion profile turns out to be a
trapezoidal velocity profile [26]. The free-vibration conditions
B. Pole-Zero Cancellation for trapezoidal velocity profile are obtained by substituting
β = 0 and γ = 1 into C1 and C2. Then, we have
As a matter of fact, the free-vibration conditions C1 and C2
are to have the pole-zero cancellations in (8). Observe that the fn τm = m. (18)
HA et al.: ROBUST ZERO PLACEMENT FOR MOTION CONTROL OF LIGHTLY DAMPED SYSTEMS 3861

TABLE I
C OMPARISON OF I NSENSITIVITIES U NDER fn τm = 1.587

as the actual natural frequency deviates from the modeled one,

the amount of the residual vibration increases. In particular, it is
noted that the motion profile tuned at a crater keeps the residual
vibration (much) lower than at a chasm when the modeling
error is less than a certain range (i.e., 16% in this case).
Considering that the modeling error in practice may not be so
large, tuning the motion profile to be at a crater is advantageous
for enhancing the robustness. For rigorous comparison, let us
Fig. 9. Sensitivity curves of the dimensionless residual vibration amplitude define a measure for robustness—insensitivity. The insensitivity
with respect to the variation of the natural frequency with fn τm = 1.587.
is defined as the range of normalized natural frequency in which
the dimensionless residual vibration (i.e., a∗ ) is less than a
Contrary to S-curve and AS-curve, the trapezoidal velocity tolerance of 0.05 (i.e., 5%). From Fig. 9, it is easy to have the
profile has only one free-vibration condition because the free- insensitivities as summarized in Table I. Observe that the crater
vibration condition C1 is meaningless when β = 0. Similar to condition produces the AS-curve three times more robust than
(17), the free-vibration condition for trapezoidal velocity profile the chasm does.
(18) may not be met with ease. This phenomenon can be analytically demonstrated by the
In summary, the trapezoidal velocity profile has only a robust pole-zero cancellation. Contrary to a chasm having a
free-vibration condition determined by the characteristics of zero, a crater condition places double zeros at the pole of the
the system and the actuator specification. Contrary to trape- system. Thus, at a crater, it holds that
zoidal velocity profile, S-curve and AS-curve have two free-
vibration conditions. In particular, AS-curve provides more s2 X1 (s) s=sp
=0
flexibility in adjusting the motion profile to minimize the
d
residual vibration than the others due to the additional tuning s2 X1 (s) = 0. (19)
parameters. ds s=sp

In fact, (19) implies that s2 X1 (s) is zero not only at s = sp but

V. ROBUSTNESS AGAINST M ODELING E RRORS also at approximately zero around s = sp . That is, observe that

The free-vibration conditions C1 and C2 depend on the d 2

s2 X1 (s+Δs) s=sp ∼
= s2 X1 (s) s=sp + s X1 (s) ·Δs.
system parameter (i.e., fn ). If there is a modeling error which is ds s=sp
inevitable in practice, then the free-vibration conditions will not (20)
result in zero vibration. Therefore, the motion control designers
Hence, (19) guarantees that s2 X1 (s + Δs)|s=sp ∼ = 0 for small
have to consider not only minimizing the residual vibration but
Δs (i.e., around s = sp ), which means that the pole is robustly
also the robustness against modeling errors.
canceled out by the zeros even though there exists a modeling
In general, slower movement (i.e., with large values of β and
error to some extent.
γ) induces smaller residual vibration. As shown in Fig. 5, as
In the case of trapezoidal velocity profile having a set of
the value of β or γ increases, the region with small residual
zeros, it is impossible to achieve the crater condition. However,
vibration gets wider. In other words, as β and γ increase, the
the S-curve and the AS-cure have two sets of zeros which may
residual vibration becomes more robust against the variation of
allow the crater condition. In particular, in the case of the AS-
the system parameter. However, it is undesirable to increase β
curve, the crater can be met with ease by adjusting two design
and γ much since this leads to slow movement.
parameters (i.e., β and γ). This is the reason why the AS-curve
The robust conditions against the modeling errors with fast
is advantageous over the other profiles in terms of robustness.
movement are investigated. The conditions satisfying both free-
The algorithm to design the AS-curve to be in a crater
vibration conditions C1 and C2 are defined as a crater. Dif-
condition is presented in the following.
ferent from the crater, the conditions satisfying only one of
Algorithm: Suppose that Vmax and Amax are fixed by the
the free-vibration conditions are defined as a chasm. As shown
actuator specification and the nominal system parameters such
in Fig. 5, compared to a chasm, a crater has wider range of
as fn and ζ are known
small residual vibration. This can be visualized by plotting a
sensitivity curve to show the dimensionless residual vibration S1. Based on Vmax and Amax , τm is calculated from (2).
with respect to the natural frequency variation. Sensitivity S2. For fast movement, choose a γ from C2 with the smallest
curves for a crater (i.e., β = 0.5 and γ = 1.26) and a chasm positive integer value of m satisfying γ ≥ 1.
(i.e., β = 0.75 and γ = 1.26) are shown in Fig. 9. Note that, S3. Based on C1, β is selected as 1/m with k = 1.
3862 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 9, SEPTEMBER 2013

Fig. 10. Experimental setup.

TABLE II
PARAMETERS FOR E XPERIMENTAL S ETUP

Fig. 11. Experimental measurements of the base movement. (Top) Accelera-

tion, (middle) velocity, and (bottom) position profiles of the base.

VI. E XPERIMENTS
In this section, the proposed tuning method is validated
through experiments with an XY stage with a flexible structure.
In practice, many of the automated machineries are built on (or
with) XY stages. For examples, XY stages are widely adopted
in industrial applications such as a crane, an inspection robot,
and chip assembly equipment [26]. For achieving high bonding
accuracy with highly integrated electronic components, the
residual vibration is a challenge to chip assembly equipment
with flexible tool tip [28]. Moreover, in the case of fine-pitch
assembly processes, the residual vibration of a tool tip caused
by the stage motion may induce the misalignment of specimens
and low working speed.
To verify the effectiveness of the proposed robust tuning rule
of the motion profile, an experimental system with an XY stage
is adopted as shown in Fig. 10. To realize the flexible mode in a
tool tip or an end effector of a manufacturing machine, a flexible
beam is attached to the base driven by a belt-type transmission
powered by an ac motor (HC-KFS13, MITSUBISHI Inc.).
For achieving the prescribed motion profile, a motion control
board (PCI-7345, NI Corporation) is used to generate driving Fig. 12. (Top) Acceleration profile of the base and (middle and bottom)
induced vibration of the flexible beam.
signals with a sampling period of 3 ms. The ac motor is con-
trolled by a servomotor driver (MR-J2S-10A, MITSUBISHI TABLE III
Inc.) using a proportional–integral–derivative control with an C OMPARISON OF S ETTLING T IMES
encoder. The physical parameters used for the experiments are
summarized in Table II. To test the robustness of the proposed
method, an additional mass is attached to vary the natural
frequency of the flexible beam from 15.87 to 14.28 Hz. The
damping ratio is 0.015. Considering the actuator specification,
the maximum velocity and maximum acceleration are chosen displacement sensor (LK-G505, KEYENCE Corporation) and
as 0.14 m/s and 1.4 m/s2 , respectively. The target distance is an accelerometer (MMA-7260Q, FREESCALE Inc.), respec-
270 mm. The absolute position of the base and the relative tively. The measured sensor signals are recorded by the data
vibration of the flexible beam are measured using a laser acquisition board (DAQ-6251, NI Corporation).
HA et al.: ROBUST ZERO PLACEMENT FOR MOTION CONTROL OF LIGHTLY DAMPED SYSTEMS 3863

Fig. 13. Acceleration profile of the base and induced vibration of the flexible beam: AS-curve tuned at (a) a crater and (b) a chasm with/without modeling error.
(a) AS-curve tuned as crater (β = 0.5; γ = 1.26). (b) AS-curve tuned as chasm (β = 0.75; γ = 1.26).

TABLE IV
C OMPARISON OF S ETTLING T IMES

Using the described system, we performed the experiments The settling times are measured and summarized in Table IV.
with (β, γ) = (0.5, 1) and (β, γ) = (0.5, 1.26). Note that the The increments of settling time due to the modeling error
case of (β, γ) = (0.5, 1) corresponds to a conventional S-curve. for the motion profile tuned at a crater and a chasm are 22.01%
Moreover, as shown in Fig. 5, the case of (β, γ) = (0.5, 1.26) and 43.44%, respectively. It is obvious that the motion profile
is the AS-curve tuned for placing double zeros on the pole (i.e., tuned at a crater is desirable in the presence of modeling error.
crater). As can be seen in the Fig. 11, the velocity profiles (of As a remark, through the experiments, we observed that
the base) actually obtained are close to the nominal design but the proposed approach is an effective and efficient method for
not exactly due to mechanical nonlinearities such as friction and reducing the residual vibration when the damping ratio is very
the belt fluctuation. These nonlinearities may cause additional small. Moreover, the analysis presents a unified framework that
vibration during motion. Fig. 12 shows the acceleration profile can suggest a systematic design guideline for generating the
of the base ẍ1 and the induced vibration of the flexible beam motion profiles, including AS-curve, S-curve, and trapezoidal
ẍ2 . Observe that the residual vibration of the AS-curve (in velocity profile.
a crater condition) significantly reduces, compared with the
S-curve case. For comparison, the settling times within which VII. C ONCLUSION
the relative acceleration of the flexible beam gets smaller than
a tolerance (i.e., ±0.1 m/s2 ) are measured and summarized in In this paper, a unified framework to minimize the resid-
Table III. The settling time of the AS-curve at a crater reduces ual vibration of the motion profiles, including AS-curve,
by 28.7% compared with the S-curve. Note that the perfect S-curve, and trapezoidal velocity profile, has been derived using
attenuation of the residual vibration under the proposed tuning Laplace- and time-domain approaches. The proposed frame-
conditions could not be attained because of the system nonlin- work can illustrate the physical characteristics of the motion
earities. Nevertheless, it turns out that the proposed approach profiles analytically. In particular, the robust tuning rule against
can improve the motion control performance effectively. the modeling error was newly discussed using the robust pole-
For demonstrating an advantage of robust pole-zero cancel- zero cancellation concept, which is effective in practice having
lation, we performed the experiments with the motion profiles modeling errors inevitably. Through the experiments, we have
tuned at a crater and a chasm in the presence of 10% modeling demonstrated the effectiveness of the proposed approach.
error. As described in the previous section, placing double zeros The proposed approach is expected to suggest an efficient
on the pole (i.e., to be a crater) is more robust against modeling design guideline for generating motion control profiles with the
errors than placing a single zero on the pole (i.e., to be a chasm). enhanced motion control performance.
As shown in Fig. 13, in the absence of modeling error, the
residual vibration is minimized in both cases. However, as the R EFERENCES
natural frequency varies by 10%, the amount of the residual [1] T. Atsumi and W. C. Messner, “Optimization of head-positioning control
vibration increases. Note that the increment of the residual in a hard disk drive using the RBode plot,” IEEE Trans. Ind. Electron,
vol. 59, no. 1, pp. 521–529, Jan. 2012.
vibration at a crater is much smaller than that at a chasm. It [2] T. R. Kurfess, Robotics and Automation Handbook. Boca Raton, FL:
can be also compared with ease by measuring the setting times. CRC Press, 2005.
3864 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 60, NO. 9, SEPTEMBER 2013

[3] M. Brady, J. M. Hollerbach, T. L. Johnson, T. L. Perez, and [26] K-.H. Rew and K-.S. Kim, “A closed-form solution to asymmetric motion
M. T. Mason, Robot Motion: Planning and Control. Cambridge, MA: profile allowing acceleration manipulation,” IEEE Trans. Ind. Electron,
MIT Press, 1979. vol. 57, no. 7, pp. 2499–2506, Jul. 2010.
[4] PMAC Manual, Delta Tau Data Systems, Inc., Chatsworth, CA, 1996. [27] C-.W. Ha, K-.H. Rew, and K-.S. Kim, “Formulating a Laplace domain
[5] K. D. Nguyen, I. M. Chen, and T. C. Ng, “Planning algorithms for S-curve approach for tuning motion profile,” in IEEE/ASME Int. Conf. Adv. Intell.
trajectories,” in Proc. IEEE/ASME Int. Conf. Adv. Intell. Mechatronics, Mechatronics, Budapest, Hungary, 2011, pp. 356–360.
ETH Zurich, Switzerland, 2007, pp. 1–6. [28] N. C. Singer, Residual Vibration Reduction in Computer Controlled
[6] J. Wu, Z. Xiong, K.-M. Lee, and H. Ding, “High-acceleration precision Machines. Cambridge, MA: MIT Press, 1989.
point-to-point motion control with look-ahead properties,” IEEE Trans.
Ind. Electron, vol. 58, no. 9, pp. 4343–4352, Sep. 2011.
[7] H. Ding and J. Wu, “Point-to-point motion control for a high-acceleration
positioning table via cascaded learning schemes,” IEEE Trans. Ind.
Electron, vol. 54, no. 5, pp. 2735–2744, Oct. 2007.
[8] Y-.S. Kung, C-.C. Huang, and M-.H. Tasi, “FPGA realization of an
Chang-Wan Ha (S’09) received the B.S. degree in
adaptive fuzzy controller for PMLSM drive,” IEEE Trans. Ind. Electron,
vol. 56, no. 8, pp. 2923–2932, Aug. 2009. mechanical engineering from the Department of Me-
chanical and Control Engineering, Handong Global
[9] C. S. Chen and A. C. Lee, “Design of acceleration/deceleration profiles in
University, Pohang, Korea, in 2008 and the M.S.
motion control based on digital FIR filters,” Int. J. Mach. Tools Manuf.,
degree in mechanical engineering from the Depart-
vol. 38, no. 7, pp. 799–825, Jul. 1998.
ment of Mechanical Engineering, Korea Advanced
[10] K. Erkokmaz and Y. Altintas, “High speed CNC system design. Part 1:
Institute of Science and Technology, Daejeon, Korea,
Jerk limited trajectory generation and quintic spline interpolation,” Int. J.
in 2010, where he is currently working toward the
Mach. Tools Manuf., vol. 41, no. 9, pp. 1323–1345, Jul. 2001.
[11] R. C. Hong and Y. F. Chang, “Parametric curve machining of a CNC Ph.D. degree.
His research interests include motion control, in-
milling EDM,” Int. J. Mach. Tools Manuf., vol. 45, no. 7/8, pp. 941–948,
put shaping, bonding technology, and semiconductor
Jun. 2005.
manufacturing equipment.
[12] B.-H. Chang and Y. Hori, “Trajectory design considering derivative of
jerk for head-positioning of disk drive system with mechanical vibra-
tion,” IEEE/ASME Trans. Mechatronics, vol. 11, no. 3, pp. 273–279,
Jun. 2006.
[13] P. H. Meckl and P. B. Arestides, “Optimized S-curve motion profiles for
minimum residual vibration,” in Proc. Amer. Control Conf., Philadelphia,
PA, 1998, pp. 2627–2631.
[14] P. H. Meckl and R. Kinceler, “Robust motion control of flexible systems Keun-Ho Rew (M’10) received the B.S. and M.S.
using feedforward forcing functions,” IEEE T. Contr. Syst. T., vol. 2, no. 3, degrees in mechanical engineering and the Ph.D.
pp. 245–254, Sep. 1994. degree in aerospace engineering from the Korea
[15] R. Zanasi, C. G. L. Bianco, and A. Tonielli, “Nonlinear filters for Advanced Institute of Science and Technology,
the generation of smooth trajectories,” Automatica, vol. 36, no. 3, pp. 439– Daejeon, in 1994, 1996, and 2001, respectively.
448, Mar. 2000. He was a Senior Engineer with Mirae Industry
[16] C. G. L. Bianco and R. Zanasi, “Smooth profile generation for a tile Company, Ltd., Cheonan, Korea, from 2001 to 2003
printing machine,” IEEE Trans. Ind. Electron, vol. 50, no. 3, pp. 471–477, and a Chief Engineer with Fine D&C Company, Ltd.,
Jun. 2003. from 2003 to 2005. In 2005, he joined the Depart-
[17] W. M. Tao, M. J. Zhang, O. Ma, and X. P. Yun, “Industrial robot ment of Mechanical Engineering, Hoseo University,
track modeling and vibration suppression,” Ind. Robot., vol. 34, no. 4, Asan, Korea, as a Faculty Member, where he has
pp. 317–325, Jun. 2007. been with the Department of Robotics Engineering since 2007. His research
[18] C. T. Freeman, Z. Cai, E. Rogers, and P. L. Lewin, “Iterative learning interests include motion control, robotics, semiconductor manufacturing equip-
control for multiple point-to-point tracking application,” IEEE Trans. Ind. ment, and digital image processing techniques.
Electron, vol. 19, no. 3, pp. 590–600, May 2011.
[19] H. Li, M. D. Le, Z. M. Gong, and W. Lin, “Motion profile design to
reduce residual vibration of high-speed positioning stages,” IEEE/ASME
T. Mech., vol. 14, no. 2, pp. 264–269, Apr. 2009.
[20] Y. J. Shin and P. H. Meckl, “Application of combined feedforward
and feedback controller with shaped input to benchmark problem,”
J. Dyn. Syst., Meas., Control, vol. 132, no. 2, pp. 021001-1–021001-8, Kyung-Soo Kim (M’00) received the B.S., M.S.,
Mar. 2010. and Ph.D. degrees in mechanical engineering
[21] A. Abe, “Trajectory planning for flexible Cartesian robot manipulator by from the Department of Mechanical Engineering,
using artificial neural network: Numerical simulation and experimental Korea Advanced Institute of Science and Technology
verification,” Robotica, vol. 29, no. 5, pp. 797–804, Sep. 2010. (KAIST), Daejeon, Korea, in 1993, 1995, and 1999,
[22] D. M. Tsay and C. F. Lin, “Asymmetrical inputs for minimizing residual respectively.
response,” in Proc. IEEE Int. Conf. Mechatronics, Taipei, Taiwan, 2005, He was a Chief Researcher with LG Electronics,
pp. 235–240. Inc., from 1999 to 2003 and as a DVD Group Man-
[23] K-.H. Rew and K-.S. Kim, “Using asymmetric S-curve profile for fast and ager with STMicroelectronics Company, Ltd., from
vibrationless motion,” in Proc. Int. Conf. Control, Autom. Syst., Seoul, 2003 to 2005. In 2005, he joined the Department of
Korea, 2007, pp. 500–504. Mechanical Engineering, Korea Polytechnic Univer-
[24] C-. W. Ha, K-. H. Rew, and K-. S. Kim, “A complete solution to asym- sity, Seoul, Korea, as a Faculty Member. Since 2007, he has been with the
metric S-curve motion profile: Theory & experiments,” in Proc. Int. Conf. Department of Mechanical Engineering, KAIST. He serves as an Associate
Control, Autom. Syst., Seoul, Korea, 2008, pp. 2845–2849. Editor of Automatica and Journal of Mechanical Science and Technology. His
[25] K-.H. Rew, C-.W. Ha, and K-.S. Kim, “A practically efficient method for research interests include digital system design for controlled mechatronics,
motion control based on asymmetric velocity profile,” Int. J. Mach. Tools sensor and actuator design, robot manipulator design, and control theories such
Manuf., vol. 49, no. 7/8, pp. 678–682, Jun. 2009. as robust control and sliding mode control.