Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO.

6, DECEMBER 2007

3413

An Analysis of Parameter Variations of


Disturbance Observer for Motion Control
Hideyuki Kobayashi, Seiichiro Katsura, Member, IEEE, and Kouhei Ohnishi, Fellow, IEEE

AbstractThis paper analyzes a disturbance observer with a


focus on parameter variations. The parameter variations are the
inertia variation and the variation of torque coefficient. Conventionally, their nominal values are designed not as control parameters but the same values as the actual ones. However, disturbance
observer is able to include the effect of phase compensation by
properly selecting the nominal model parameters. In this paper,
these nominal model parameters are actively designed to achieve
the phase compensation. This paper proposes the design method
of observer to achieve phase compensation based on disturbance
observer. Compared with implementing phase compensator, the
control system is simple and easy to design. The selection of
these parameters has some restriction. When the nominal model
parameters are much separated from the actual ones, the stability
of the control system deteriorates. This paper focuses on the
inertia variation and neglects the variation of torque coefficient.
Furthermore, the stability of position and force control systems
is analyzed. The validity of the proposed design method and the
effect of phase compensation are verified by some experimental
results.
Index TermsAcceleration control, disturbance observer, force
control, motion control, parameter variations, phase compensation, position control, robustness.

I. I NTRODUCTION

OTION CONTROL has been developed in recent


decades. The first IEEE International Workshop on Advanced Motion Control held in 1990 pointed out the importance
of physical interpretation of motion control. The software servoing technology is now common in machine tools, robotics,
and mechatronics. It has been intensively developed for the
numerical-control machines. Conventional industrial machines
have aimed at disturbance rejection. Basically, the high gain
feedback adopted in such a position controller tends to cause
the so-called woodpecker phenomena during contact with hard
environment.

Manuscript received March 1, 2006; revised July 10, 2007. This work was
supported by the Ministry of Education, Culture, Sports, Science and Technology of Japan under Grant-in-Aid for Scientific Research (A), 17206027, 2005.
This paper was presented in part at the 31st IEEE Annual Conference of the
IEEE Industrial Electronics Society, Raleigh, USA, November 6-10, 2005.
H. Kobayashi is with the System Development Laboratory, Hitachi, Ltd.,
Yokohama 244-0817, Japan (e-mail: hide@2005.jukuin.keio.ac.jp).
S. Katsura is with the Department of Electrical Engineering, Nagaoka
University of Technology, Nagaoka 940-2188, Japan (e-mail: katsura@vos.
nagaokaut.ac.jp).
K. Ohnishi is with the Department of System Design Engineering, Keio
University, Yokohama 223-8522, Japan (e-mail: ohnishi@sd.keio.ac.jp).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIE.2007.905948

In order to stably contact with the environment, many researches have been developed. The basic strategy to contact
with the environment is stiffness control [1], which corresponds
to proportional-derivative control with gravity compensation.
Active compliance can be provided by using either hybrid
position/force control [2] or impedance control [3][5]. Contact
transition control via acceleration feedback [6] and neural network control scheme for force control [7] have been proposed.
In this way, many researchers have paid attention to develop
novel force control architectures.
The ability which is requested of machines is changing
recently. In the past, machines only worked in closed environments such as a factory or plant. However, recent machines are
being planned to work in an open environment, which dynamically changes. In order to make machines that are able to work
in such environments, haptics is the key technology in addition
to vision and speech. Haptics is a technology for touching an
object. Teleoperation [8][12] and virtual reality [13] are some
of the expected applications of haptics. Shimono et al. [14]
proposed the haptic database based on bilateral control. The
proposed control system abstracts the haptic information of an
environment and reproduces the haptic information bilaterally.
Katsura and Ohnishi [15] proposed the haptic training system
by multilateral control. It attains the law of action and reaction by three remote robots and realizes the haptic training
system.
The robustness of motion systems is essential to attain such
high performance as the haptic motion control. A good candidate for the attainment of robust control is a disturbance
observer [16], [17]. Disturbance observer is a basic technology
for the disturbance suppression and the attainment of robust
acceleration control. The disturbance observer identifies the
total mechanical load torque and parameter variations. The
parameter variations are the inertia variation and the variation
of torque coefficient. The feedback of the estimated disturbances makes the control system nominal. Conventionally, the
nominal values of inertia and torque coefficients are designed
not as control parameters but the same values as the actual
ones. However, the disturbance observer is able to include
the effect of phase compensation by properly selecting the
nominal model parameters. Compared with implementing a
conventional phase lead-lag compensator, the control system
is simple and easy to design. Katsura and Ohnishi [18] proposed the absolute stabilization of a multimass resonant system
by phase compensation based on disturbance observer. The
proposed control focuses on nominal inertia and attains the
phase lead compensation by properly selecting the nominal
inertia.

0278-0046/$25.00 2007 IEEE

3414

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 6, DECEMBER 2007

In this paper, the nominal model parameters are actively designed to achieve the phase compensation. This paper proposes
the design method of observer to achieve phase compensation
based on disturbance observer. The selection of these nominal
model parameters has some restriction. When the nominal
model parameters are much separated from the actual ones,
the stability of the control system deteriorates. Murakami and
Ohnishi [19] researched the stability of a position control
system when the nominal inertia changes. Bertoluzzo et al. [20]
also researched the stability in discrete-time system. However,
these researches did not focus on the bandwidth for disturbance
suppression. It is commonly misconceived that the bandwidth
for disturbance suppression is equal to the cutoff frequency of
disturbance observer. The main contribution of this paper is to
show that the bandwidth for disturbance suppression depends
not only on the cutoff frequency of disturbance observer but
also on the parameter variations. The bandwidth for disturbance
suppression should be fixed when the nominal model parameters are actively designed. The stability under the parameter
variations should also be analyzed with the bandwidth for
disturbance suppression fixed. Since the variation of torque
coefficient is small, this paper focuses on the inertia variation
and neglects the variation of torque coefficient. Furthermore,
the stability of position control system and force control system
is analyzed.
This paper is organized as follows. In the following section,
control stiffness and disturbance observer are introduced as
basic skills of motion control. In this section, it will be shown
that the cutoff frequency of disturbance suppression depends
on the parameter variations. This is the main contribution of this
paper. It will also be shown that a disturbance observer includes
the effect of phase compensation. In Section III, the method
to design the cutoff frequency of the disturbance observer and
reaction torque observer is proposed. In the proposed design
method, the nominal model parameters are actively chosen
to achieve the phase compensation based on the disturbance
observer. When the nominal model parameters are much
separated from the actual ones, the stability of the control system deteriorates. The stability of position and force
control systems is analyzed in Section IV. Then, the validity of the proposed design method and the effect of
phase compensation are verified by some experimental results
shown in Section V. Finally, this paper is summarized in
Section VI.

II. M OTION C ONTROL


A. Target of Motion Control
Recent machines are required to have an ability to make
contact with an unknown environment. In motion control under an open environment, robots are required to adapt to the
interference from other systems and the environment where it
dynamically changes. A more sophisticated ability in motion
control is necessary for compliant contact with the environment. Conventional motion control is not always suitable for
future applications due to the lack of adaptive capability to
the environment. A motion controller in an open environment

will require various stiffnesses corresponding to the task. The


control stiffness is defined in the following equation:
=

f
x

(1)

where denotes a control stiffness, f denotes a generated force,


and x denotes a position. The ideal position control should have
an infinite stiffness. In contrast, the ideal force control should
have zero stiffness. Modern motion systems require an ability
to be compliant with the contact environment. The motion
control should take plural modes of environment into account.
The environmental information is a key to be compliant to
an unknown environment. Motion control, taking environment
into account, will make various machines more adaptive and
versatile.
A high-performance motion controller is a requisite for more
skillful and dexterous motion. In order to attain such high
performance, motion systems should be robust against a load
change and parameter variations. Sabanovic [21] mentioned
that an acceleration controller makes a motion system robust
by analyzing the sliding mode control. Hence, the dimension
of reference value, which is input into the acceleration controller, must be acceleration. From the point of view of control
stiffness, the position and force controls conflict. However, it is
possible to uniformly deal with the two controls in acceleration
dimension
d2
x
dt2
f
x
=
m

x
=

(2)
(3)

where x
denotes acceleration, and m denotes mass. Therefore, the acceleration controller is able to uniformly deal with
the position and force controls. In robust motion control, the
acceleration reference generator makes the acceleration reference value for the attainment of desired control stiffness.
The acceleration reference value is accurately attained by the
acceleration controller. Finally, the desired control stiffness is
achieved.
B. Acceleration Control by Disturbance Observer
A disturbance observer, which was proposed by Ohishi et al.
[16], [17], is a good candidate for an acceleration controller. A
disturbance observer identifies the total mechanical load torque
and parameter changes. In other words, the identification of
disturbance torque is essential for motion control robustness to
realize various applications. This section introduces the disturbance observer as an acceleration controller. The disturbance
observer is designed to cancel the disturbance torque as quickly
as possible.
The dynamic equation of motion system is represented as
follows:
J

d
= a l
dt
a = Kt Ia

(4)
(5)

KOBAYASHI et al.: ANALYSIS OF PARAMETER VARIATIONS OF DISTURBANCE OBSERVER FOR MOTION CONTROL

3415

where
J
inertia;
a generated torque;
load torque;
l
Kt torque coefficient of electric motor;
Ia torque current.
The parameters in (4) and (5) are inertia and torque coefficient. The inertia will change according to the mechanical
configuration of motion system. The torque coefficient will vary
according to the rotor position of electric motor due to irregular
distribution of magnetic flux on the surface of rotor
J = Jn + J
Kt = Ktn + Kt

(6)
(7)

Fig. 1. Block diagram of the disturbance observer.

where
nominal inertia;
Jn
J
inertia variation;
Ktn nominal torque coefficient;
Kt variation of torque coefficient.
By substituting (6) and (7) into (4) and (5), the following
equation is obtained:
Jn res = Ktn Ia dis .

(8)

Here, the disturbance torque dis is represented as follows:


dis = (J Jn )res + (Ktn Kt )Ia
+ Fc + Dres + reac

(9)

= J res Kt Ia + l

(10)

l = Fc + Dres + reac .

(11)

In (9), the first term is the torque pulsations due to the inertia
variation. The second term is the torque pulsations due to
the variation of torque coefficient. The third and fourth terms
denote the Coulomb and the viscous friction, respectively. The
last term is the reaction torque caused by external torque.
Equation (8) can be transformed into
dis = Ktn Ia Jn res .

(12)

The estimated disturbance torque is obtained from the velocity


response res and the torque current Ia , as shown in Fig. 1.
Equation (13) shows that the disturbance torque is estimated
through the first-order low-pass filter
dis =

gdis
dis
s + gdis

(13)

where gdis denotes the cutoff frequency of the low-pass filter.


The estimated disturbance torque is used for the realization of
robust motion control. By attaching the unity gain of nominal
inertia and torque coefficient, the robust motion controller
makes a motion system an acceleration control system, as
shown in Fig. 2
Iaref =

Jn ref
.
Ktn

(14)

Fig. 2. Acceleration control system attained by the disturbance observer.

As shown in Fig. 2, the effect of the load torque is represented


as a transfer function Gs [22]
Gs =

gdis
=

s + gdis

(15)

Kt Jn
gdis = gdis
Ktn J

(16)

Kt Jn
.
Ktn J

(17)

Gs represents a sensitivity showing how the load torque influences the motion system and is called a sensitivity function. As
shown in (16), the cutoff frequency of disturbance suppression

depends on the inertia variation and the variation of torque


gdis
coefficient.
Moreover, the disturbance observer includes the effect of
phase compensation
Gc =

s + gdis
s + gdis

> 1 phase lead


0<<1

phase lag.

(18)

Compared with the inertia variation, the variation of torque


coefficient is small. By assuming that the variation of torque

3416

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 6, DECEMBER 2007

Furthermore, the narrow bandwidth of a force sensor has a big


influence on the force control system [25]. The bandwidth of
a reaction torque observer is much wider than that of a force
sensor [26]. By sensing with a wide bandwidth, the ability of
a force control system is drastically improved. Therefore, the
reaction torque observer is more effective than a force sensor.
III. D ESIGN M ETHOD OF O BSERVER
In this section, the method of designing the cutoff frequency
of disturbance observer gdis and reaction torque observer greac
is introduced. As mentioned in the previous section, this paper
neglects the variation of torque coefficient and focuses on the
inertia variation.

First, the cutoff frequency of disturbance suppression gdis


is decided according to the specifications of motor, sampling
period, and so on

Fig. 3. Block diagram of the reaction torque observer.

coefficient is zero, the aforementioned inequality is represented


as follows:
Jn > J

phase lead

0 < Jn < J

phase lag.

When the nominal inertia is much separated from the actual


one, the stability of the control system deteriorates. A designer
can freely select the nominal inertia as long as the stability
of the control system is kept. Thus, the system, which is the
implemented disturbance observer, is able to include the effect
of phase compensation without a compensator.
The disturbance observer is also effective for the parameter
identification in a mechanical system. The parameters are the
friction, the inertia variation, and the variation of torque coefficient. By subtracting the torque due to the identified parameters
from the disturbance torque, it is possible to estimate the
reaction torque applied to the system


reac = dis Fc + Dres + J res Kt Ia . (19)
It is necessary and important to identify their parameters as
precisely as possible. This process is called a reaction torque
observer [23]. The block diagram of the reaction torque observer is shown in Fig. 3. The reaction torque observer can
calculate the external torque that is applied to a motor without
any sensors. Equation (20) shows that the reaction torque is
estimated through the first-order low-pass filter
reac =

greac
reac
s + greac

gdis
=G

where G is constant.
Second, the cutoff frequency of torque estimation greac is
set to the same value as the cutoff frequency of disturbance
suppression
greac = G.

(22)

The bandwidth for torque estimation is hoped to be as large


as possible. However, it is no use estimating torque beyond
the bandwidth for disturbance suppression. Hence, the cutoff
frequency of disturbance suppression and the cutoff frequency
of torque estimation should be set to the same value.
Third, nominal inertia Jn is selected while taking the
stability of a control system into account. As shown in (18),
the system, which is the implemented disturbance observer, is
able to include the effect of phase compensation by properly
selecting Jn
s + G
.
(23)
Gc =
s+G
The Bode diagram of the phase compensation is shown in
Fig. 4. The selection of the nominal inertia has some restriction.
The stability of the control system deteriorates when the nominal inertia is much separated from the actual one. It is possible
to design the nominal inertia while keeping the stability of the
control system.
Lastly, the cutoff frequency of disturbance observer gdis is
designed as follows:

(20)

where greac denotes the cutoff frequency of the low-pass filter.

, the
Unlike the cutoff frequency of disturbance suppression gdis
cutoff frequency of torque estimation greac does not depend on
other parameters.
In this paper, the reaction torque observer is selected to detect
the reaction torque from environment. Although a force sensor
is generally utilized to detect the reaction force, a force sensor
has so many problems. Since a force sensor detects the strain
of a strain gauge, it is difficult to detect the reaction force
when the environment is distorted at the time of contact [24].

(21)

gdis =

J
1
G = G.
Jn

(24)

Thereby, the bandwidth for disturbance suppression is fixed


even if the nominal inertia is set to the separate value from the
actual one
Jn
gdis = gdis
J
1
= G = G.

=
gdis

This is the order of observer design.

(25)

KOBAYASHI et al.: ANALYSIS OF PARAMETER VARIATIONS OF DISTURBANCE OBSERVER FOR MOTION CONTROL

Fig. 4.

3417

Bode diagram of phase compensation. (a) Phase lead (G = 500, = 2). (b) Phase lag (G = 500, = 0.5).
TABLE I
PARAMETERS FOR STABILITY ANALYSIS

Fig. 5. Block diagram of the position control system.

command cmd to the angle response res is represented as


follows:
IV. S TABILITY A NALYSIS
As mentioned in the previous section, the disturbance observer includes the effect of phase compensation by properly
selecting nominal model parameters. However, the stability of
the control system deteriorates when the nominal values are set
to much different values from the actual ones. This paper analyzes the stability of the position control system and the force
control system as a basic control system of motion control.
As in the previous section, the variation of torque coefficient
is neglected, and the inertia variation is focused on. For the
sake of simplicity, this paper analyzes about one-DOF system.
The control parameters for stability analysis are shown in
Table I.
In this section, the root locus for inertia variation is drawn
to analyze the stability of the control system. The shape of
root locus changes according to the values of inertia, the cutoff
frequency of disturbance suppression and torque estimation,
and so on. It is possible to solve the zeros but impossible to
solve the poles of root locus, including the character variables.
Hence, the root locus whose parameters are shown in Table I is
drawn in this section.

A. Position Control System


The block diagram of the position control system is shown in
Fig. 5. From this figure, the transfer function from the angle

) (Kp + Kv s)
(s + gdis
res
= 2

) (K + K s) .
cmd

s (s + gdis ) + (s + gdis
p
v

(26)

From the characteristic equation, the following equation is


obtained:
1+

s(Kv s + Kp )
s2 + K g s + K g = 0.
s3 + gdis
v dis
p dis

(27)

The root locus for inertia variation is shown in Fig. 6. Here,


zeros and poles are calculated as follows:
zeros = (20.0, 0.0)

(28)

poles = (914.4, 51.8, 33.8).

(29)

This result shows that the system is constantly stable even if the
nominal inertia is much separated from the actual one.
B. Force Control System
The block diagram of force control system is shown in
Fig. 7. When a machine contacts with the environment, the
reaction torque loop appears in a control system. The reaction
torque loop disappears when the machine gets away from the
environment. Generally, the environment is represented as a
spring and damper model
Zenv = Kenv + Denv s

(30)

3418

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 6, DECEMBER 2007

Fig. 8.

Fig. 6. Root locus for inertia variation (position control system).

Root locus for the inertia variation (force control system).

The root locus for inertia variation is shown in Fig. 8. Here,


zeros and poles are calculated as follows:
zeros = (1000.0, 0.0)
poles = (1009.9, 1000, 0.05 j99.5).

(37)
(38)

This result shows that the system is constantly stable even


if the nominal inertia is far separated from the actual one.
Moreover, this result shows that it is possible to place the
dominant root to further left by properly selecting the inertia
variation. When the inertia variation is 0.5, 1.0, and 2.0, roots
are calculated as follows:
roots( = 0.5) = (1000.0, 1005.0, 2.5 j99.7)
roots( = 1.0) = (1000.0, 1000.0, 5.0 j99.9)
roots( = 2.0) = (1000.0, 989.9, 10.1 j100.0).

Fig. 7. Block diagram of the force control system.

where
Zenv
Kenv
Denv

environmental impedance;
spring coefficient;
viscosity coefficient.

This means that the phase compensation due to the inertia


variation is effective for stability improvement.

The transfer function from the torque command cmd to the


estimated reaction torque reac is represented as follows:
reac
N1 (s)
=
cmd

D1 (s)

(31)

where

N1 (s) = C
(32)
f J 2(s + gdis ) (Kenv + Denv s)(s +
 greac )
D1 (s) = Js (s + gdis ) + s(Kenv + Denv s) (s + greac )

) (Kenv + Denv s).


(33)
+ Cf Jgreac (s + gdis

From the characteristic equation, the following equation is


obtained:
1+

N2 (s)
=0
D2 (s)

(34)

N2 (s) = Cf Jgreac (Kenv + Denv s)s


(35)
 3


2
D2 (s) = Js + (Jgdis + Denv ) s + Kenv s

(s + greac ) + Cf Jgreac gdis

(Kenv + Denv s).

(39)
(40)
(41)

(36)

V. E XPERIMENT
In this section, the validity of the proposed design method
and the effect of phase compensation are demonstrated by some
experimental results. First, it is verified from the results of
the position control system that the bandwidth for disturbance
suppression depends on parameter variation, as shown in (16).
Second, the effect of phase compensation is verified from the
results of the force control system.
A. Experimental Setup
Experiment is performed with a rotational manipulator
shown in Fig. 9. The overview of experiment setup is shown in
Fig. 10. Actuator is an ac servomotor that is made in Yaskawa
Electric Corporation. The power amplifier is II series. The
specifications of the ac motor are shown in Table II. Position
response is measured by the encoder. Velocity response is
calculated by the differentiation of position response through
the first-order low-pass filter
res =

gpd
sres
s + gpd

(42)

KOBAYASHI et al.: ANALYSIS OF PARAMETER VARIATIONS OF DISTURBANCE OBSERVER FOR MOTION CONTROL

Fig. 9.

One-DOF rotational manipulator.

Fig. 10. Overview of the experimental setup.


TABLE II
SPECIFICATIONS OF AC SERVOMOTOR

TABLE III
PARAMETERS FOR EXPERIMENT

3419

why the responses of (a) and (c) got worse is related to the
bandwidth. Since the bandwidth for disturbance suppression is
represented as (16), the bandwidth of (a) is smaller than that
of (b). In addition, the bandwidth of (c) is larger than that of
(b). Hence, the response of (a) has an overshoot and longer
settling time. The response of (c) includes some noise. On the
other hand, Fig. 11(d)(f) shows the position responses when
the cutoff frequency of the disturbance observer is set to (24).
In this case, the bandwidth for disturbance suppression is fixed.
The responses of (d)(f) follow the position command quickly
and precisely. From the comparison between Fig. 11(a)(c)
and (d)(f), it is verified that the bandwidth for disturbance
suppression depends on the inertia variation.
The experimental results of the force control system are
shown in Fig. 12. Fig. 12(a)(c) are the position responses and
Fig. 12(d)(f) are the force responses. The force command is
1 Nm. The environment is expanded polystyrene. The bandwidth for disturbance suppression is fixed by selecting the
cutoff frequency of the disturbance observer as (24). The
bandwidth for torque estimation is set to the same value as
the bandwidth for disturbance suppression. The responses of
(a) and (d) include the phase lag compensation based on the
disturbance observer. It shows that the system could not stably
contact with the environment. When an end-effector contacts
with the environment, a reaction torque loop appears in a
control system. Then, the system becomes oscillatory, and the
end-effector is repulsed from the environment. When the endeffector gets away from the environment, the reaction torque
loop disappears, and the system becomes stable. Furthermore,
the end-effector falls down and contacts with the environment again. This process was repeated again and again in
the responses of (a) and (d). Hence, stable contact with the
environment could not be attained in this case. On the contrary,
the responses of (b), (c), (e), and (f) show that the system could
contact stably. The responses of (b) and (e) include no phase
compensation, and the responses of (c) and (f) include the phase
lead compensation based on disturbance observer. From the
responses during contact time, it is proved that the response
of (c) was able to contact more stably than the response of (b).
This means that the phase compensation based on disturbance
observer makes the force control system more stable.
VI. C ONCLUSION

where gpd denotes the cutoff frequency of the low-pass filter.


Control software is written in C language under RT-Linux 2.4.
Sampling period is 0.1 ms. The control parameters for experiment are shown in Table III.
B. Experimental Results
The experimental results of the position control system are
shown in Fig. 11. The position command is a unit-step input.
Fig. 11(a)(c) shows the position responses when the cutoff
frequency of the disturbance observer is a constant value. In
this case, the bandwidth for disturbance suppression is not
fixed. The response of (a) has an overshoot and longer settling
time. The response of (c) includes some noise. The reason

This paper analyzed a disturbance observer with a focus on


parameter variations. The parameter variations are the inertia
variation and the variation of the torque coefficient. Conventionally, their nominal values are designed not as control
parameters but the same values as the actual ones. However,
the disturbance observer is able to include the effect of phase
compensation by properly selecting the nominal model parameters. In this paper, these nominal model parameters were
actively designed to achieve the phase compensation. This
paper proposed the design method of observer to achieve phase
compensation based on the disturbance observer. Compared
with implementing the phase compensator, the control system
is simple and easy to design. The selection of these nominal
model parameters has some restriction. When the nominal

3420

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 54, NO. 6, DECEMBER 2007

Fig. 11. Experimental results of position contral system. (a) Position response (gdis = G, = 0.1). (b) Position response (gdis = G, = 1.0).
1
1
G, = 0.1). (e) Position response (gdis =
G, = 1.0). (f) Position response
(c) Position response (gdis = G, = 1.6). (d) Position response (gdis =
1
G, = 4.0).
(gdis =

1
1
Fig. 12. Experimental results of force control system. (a) Position response (gdis =
G, = 0.5). (b) Position response (gdis =
G, = 1.0).
1
1
1
G, = 2.0). (d) Force response (gdis =
G, = 0.5). (e) Force response (gdis =
G, = 1.0). (f) Force response
(c) Position response (gdis =
1
G, = 2.0).
(gdis =

model parameters are much separated from the actual ones, the
stability of the control system deteriorates. The main contribution of this paper is to show that the bandwidth for distur-

bance suppression depends not only on the cutoff frequency of


the disturbance observer but also on the parameter variations.
The bandwidth for disturbance suppression should be fixed

KOBAYASHI et al.: ANALYSIS OF PARAMETER VARIATIONS OF DISTURBANCE OBSERVER FOR MOTION CONTROL

when the nominal model parameters are actively designed


to achieve the phase compensation based on the disturbance
observer. The stability under the parameter variations should
also be analyzed with the bandwidth for the disturbance suppression fixed. This paper focused on the inertia variation
and neglected the variation of torque coefficient. Furthermore,
the stability of position and force control systems was analyzed. The validity of the proposed design method and the
effect of phase compensation were verified by some experimental results. The proposed analysis of parameter variations
in the disturbance observer will be a basic skill for motion
control.

3421

[21] A. Sabanovic, Sliding modes in power electronics and motion control


systems, in Proc. 29th IEEE Annu. IECON, 2003, pp. 9971002.
[22] H. Kobayashi, S. Katsura, and K. Ohnishi, An analysis of parameter
variations of disturbance observer for haptic motion control, in Proc. 31st
IEEE Annu. Conf. IECON, 2005, pp. 19071912.
[23] T. Murakami, F. Yu, and K. Ohnishi, Torque sensorless control in
multidegree-of-freedom manipulator, IEEE Trans. Ind. Electron., vol. 40,
no. 2, pp. 259265, Apr. 1993.
[24] S. Katsura, Y. Matsumoto, and K. Ohnishi, Modeling of force sensing
and validation of disturbance observer for force control, IEEE Trans. Ind.
Electron., vol. 54, no. 1, pp. 530538, Feb. 2007.
[25] Y. F. Li and X. B. Chen, On the dynamic behavior of a force/torque
sensor for robots, IEEE Trans. Instrum. Meas., vol. 47, no. 1, pp. 304
308, Feb. 1998.
[26] S. Katsura, Y. Matsumoto, and K. Ohnishi, Analysis and experimental
validation of force bandwidth for force control, IEEE Trans. Ind. Electron., vol. 53, no. 3, pp. 922928, Jun. 2007.

R EFERENCES
[1] J. K. Salisbury, Active stiffness control of a manipulator in Cartesian
coordinates, in Proc. 19th IEEE CDC, 1980, pp. 95100.
[2] M. H. Raibert and J. J. Craig, Hybrid position/force control of manipulators, Trans. ASME, J. Dyn. Syst. Meas. Control, vol. 102, pp. 126133,
1981.
[3] N. Hogan, Impedance control: An approach to manipulator,
Part: ITheory, Trans. ASME J. Dyn. Syst. Meas. Control, vol. 107,
no. 1, pp. 17, Mar. 1985.
[4] N. Hogan, Impedance control: An approach to manipulator,
Part: IIImplementation, Trans. ASME J. Dyn. Syst. Meas. Control,
vol. 107, no. 1, pp. 816, Mar. 1985.
[5] N. Hogan, Impedance control: An approach to manipulator,
Part: IIIApplications, Trans. ASME J. Dyn. Syst. Meas. Control,
vol. 107, no. 1, pp. 1724, Mar. 1985.
[6] X. L. Wu, J. D. Han, and S. K. Tso, Experimental study of contact
transition control incorporating joint acceleration feedback, IEEE/ASME
Trans. Mechatronics, vol. 5, no. 3, pp. 292301, Sep. 2000.
[7] S. Jung and T. C. Hsia, Robust neural force control scheme under uncertainties in robot dynamics and unknown environment, IEEE Trans. Ind.
Electron., vol. 47, no. 2, pp. 403412, Apr. 2000.
[8] D. A. Lawrence, Stability and transparency in bilateral teleoperation,
IEEE Trans. Robot. Autom., vol. 9, no. 5, pp. 624637, Oct. 1993.
[9] Y. Yokokohji and T. Yoshikawa, Bilateral control of masterslave manipulator for ideal kinesthetic couplingFormulation and experiment,
IEEE Trans. Robot. Autom., vol. 10, no. 5, pp. 605620, Oct. 1994.
[10] S. Katsura, Y. Matsumoto, and K. Ohnishi, Realization of law of action
and reaction by multilateral control, IEEE Trans. Ind. Electron., vol. 52,
no. 5, pp. 11961205, Oct. 2005.
[11] S. Katsura, Y. Matsumoto, and K. Ohnishi, Transmission of force sensation by environment quarrier based on multilateral control, IEEE Trans.
Ind. Electron., vol. 54, no. 2, pp. 898906, Apr. 2007.
[12] Y. Matsumoto, S. Katsura, and K. Ohnishi, Dexterous manipulation
in constrained bilateral teleoperation using controlled supporting point,
IEEE Trans. Ind. Electron., vol. 54, no. 2, pp. 11131121, Apr. 2007.
[13] R. J. Adams and B. Hannaford, Stable haptic interaction with virtual
environments, IEEE Trans. Robot. Autom., vol. 15, no. 3, pp. 465474,
Jun. 1999.
[14] T. Shimono, S. Katsura, and K. Ohnishi, Abstraction and reproduction of
force sensation from real environment by bilateral control, IEEE Trans.
Ind. Electron., vol. 54, no. 2, pp. 907918, Apr. 2007.
[15] S. Katsura and K. Ohnishi, A realization of haptic training system by
multilateral control, IEEE Trans. Ind. Electron., vol. 53, no. 6, pp. 1935
1942, Dec. 2007.
[16] K. Ohishi, K. Ohnishi, and K. Miyachi, Torque-speed regulation of DC
motor based on load torque estimation, in Proc. IEEJ IPEC-TOKYO,
1983, vol. 2, pp. 12091216.
[17] K. Ohnishi, M. Shibata, and T. Murakami, Motion control for advanced
mechatronics, IEEE/ASME Trans. Mechatronics, vol. 1, no. 1, pp. 5667,
Mar. 1996.
[18] S. Katsura and K. Ohnishi, Absolute stabilization of multi-mass resonant
system by phase-lead compensator based on disturbance observer, in
Proc. 8th IEEE Int. Workshop AMC, 2006, pp. 194199.
[19] T. Murakami and K. Ohnishi, A study of stability and workspace decoupling control based on robust control in multi-degrees-of-freedom robot,
IEEJ Trans. Ind. Appl., vol. 113-D, no. 5, pp. 639646, 1993.
[20] M. Bertoluzzo, G. Buja, and E. Stampacchia, Performance analysis of
a high-bandwidth torque disturbance compensator, IEEE/ASME Trans.
Mechatronics, vol. 9, no. 4, pp. 653660, Dec. 2004.

Hideyuki Kobayashi received the B.E. degree in


system design engineering and the M.E degree in
integrated design engineering from Keio University,
Yokohama, Japan, in 2005 and 2007, respectively.
He is currently with the System Development
Laboratory, Hitachi, Ltd., Yokohama, Japan. His research interests include haptics, motion control, and
robotics.
Mr. Kobayashi received the IES Student Best
Paper Award from the IEEE Industrial Electronics
Society in 2006.

Seiichiro Katsura (S03M04) received the B.E.


degree in system design engineering and the M.E.
and Ph.D. degrees in integrated design engineering
from Keio University, Yokohama, Japan, in 2001,
2002, and 2004, respectively.
From 2003 to 2005, he was a Research Fellow
with the Japan Society for the Promotion of Science.
From 2004 to 2005, he was also a Postdoctoral
Fellow with the Keio University. Since 2005, he has
been with the Department of Electrical Engineering, Nagaoka University of Technology, Nagaoka,
Japan. His research interests include haptics, motion control, mechatronics, and
robotics.
Dr. Katsura received the Best Paper Award from the Institute of Electrical
Engineers of Japan in 2003 and the Yasujiro Niwa Outstanding Paper Award
in 2004.

Kouhei Ohnishi (S78M80SM00F01) received the B.E., M.E., and Ph.D. degrees in electrical
engineering from the University of Tokyo, Tokyo,
Japan, in 1975, 1977, and 1980, respectively.
Since 1980, he has been with the Department
of System Design Engineering, Keio University,
Yokohama, Japan. His research interests include
power electronics, mechatronics, motion control, and
robotics.
Dr. Ohnishi received Best Paper Awards from the
Institute of Electrical Engineers of Japan and the
Japan Society for Precision Engineering, and the Outstanding Paper Awards
at IECON85, IECON92, and IECON93. He also received the EPE-PEMC
Council Award and the Dr.-Ing. Eugene Mittelmann Achievement Award from
the IEEE Industrial Electronics Society in 2004.

You might also like