You are on page 1of 13

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO.

3, AUGUST 2008 185

Stability Analysis and Practical Design Procedure


of Time Delayed Control Systems With
Communication Disturbance Observer
Kenji Natori, Student Member, IEEE, Roberto Oboe, Senior Member, IEEE, and Kouhei Ohnishi, Fellow, IEEE

Abstract—In a research field of network-based control systems For the practical use of NBCSs, one of the most significant
(NBCSs), the time delay problem is one of the most significant is- issues is network-induced time delay. Time delay in NBCSs
sues. Efficient stabilization methods of time delayed control sys- deteriorates the performance and possibly leads to instability.
tems enable NBCSs to be flexibly applied to many kinds of situa-
tions. A novel time delay compensation method based on the con- Zhang et al. analyzed stability of NBCSs and discussed com-
cept of network disturbance (ND) and communication disturbance pensation of network-induced delay [2]. Then Soucek et al. con-
observer (CDOB) has been proposed. The compensation method sidered the effect of delay jitter on quality of control (QoC)
has the same effectiveness as that of the Smith predictor. In addi- [3]. About the performance issue, Lee et al. studied the per-
tion, since the method is simple and does not need time delay model formance of the quality of service (QoS)-based remote con-
or time delay measurement, it can be easily implemented to various
applications. However, the design method has not been concerned trol NBCSs with network delay [4]. For the design of NBCSs,
so far. This paper therefore presents stability analysis and studies a Chow et al. proposed gain scheduler middleware (GSM) for
practical design procedure of the time delayed control systems with simple adaptation of control systems to various delay time and
CDOB. At first, the concept of ND is introduced and the validity of applied the method to a teleoperation system [5], [6]. Yue et
the time delay compensation method is described. Then an analysis al. studied the design of robust controllers for uncertain
about the effects of parameters in control systems on stability is
conducted. Characteristics of the effects of parameters on stability NBCSs considering the effects of network-induced delay and
come out. Then we study a practical design procedure of the time data dropout [7]. In [8], NBCS was modeled as a sampled-
delayed control systems. The validity of the design procedure is val- data system with time delay and discussed about the synthesis
idated by experimental results. In the experiment, we also verify and analysis. Hwang et al. developed network-based fuzzy de-
the performance of the system in the case of time-varying delay. centralized sliding-mode control (NBFDSMC) [9]. Hu et al.
Finally, comparative study of the method to the Smith predictor is
presented. considered a novel event-driven networked predictive control
(NPC) method which does not need time delay measurement
Index Terms—Communication disturbance observer (CDOB),
[10]. Co-design of adaptive controllers and message scheduling
network-based control systems (NBCSs), network disturbance
(ND), practical design procedure, stability analysis, time delay. considering QoC was studied by Marti et al. [11].
Also in the research field of time delayed control systems, a
lot of related researches have been studied [12]. The Smith pre-
I. INTRODUCTION dictor [13] has been utilized for a long time and been extended
to various cases [14]–[17]. Besides, PID control strategy is still
UE to the rapid development of information networks, an effective method for controlling time delayed control systems
D network-based control systems (NBCSs) have been re-
ceiving attention [1]. Adoption of networks as communication
[18], [19]. Furthermore, robust control theory like control
framework is also an effective control method [20], [21]. Re-
lines in control systems yields various possibilities, smart im- cently, loop transfer recovery (LTR) is also an interesting con-
plementations, and high-speed communications of control sys- trol strategy of time delayed control systems [22].
tems. Therefore, many kinds of researches about NBCSs have In these circumstances, a novel time delay compensation
been studied so far. method based on the concept of network disturbance (ND) and
communication disturbance observer (CDOB) was proposed in
Manuscript received June 15, 2007; revised December 14, 2007, May 20,
a research field of time delayed bilateral teleoperation systems
2008, and June 25, 2008. Current version published September 10, 2008. This [23] and has been extended to time delayed control systems
work was supported in part by the Ministry of Education, Culture, Sports, Sci- [24], [25]. The compensation method has the same effective-
ence, and Technology of Japan under Grant-in-Aid for Scientific Research (S),
20226007, 2008. This paper was presented in part at the IEEE International
ness as that of the Smith predictor. Furthermore, since the
Symposium on Industrial Electronics (ISIE2007), Vigo, Spain, June 4–7, 2007. method is simple and does not need time delay model or time
Paper no. TII-07-06-0080.R3. delay measurement, it can be easily implemented to various
K. Natori and K. Ohnishi are with the Department of System Design Engi-
neering, Keio University, Yokohama 223–8522, Japan (e-mail: natori@sum.sd.
applications. However, the design issue has not been concerned
keio.ac.jp; ohnishi@sd.keio.ac.jp). so far.
R. Oboe is with the Dipartimento di Tecnica e Gestione dei Sistemi Indus- In this paper, we present stability analysis and study a prac-
triali, Universita di Padova, Vicenza 36100, Italy (e-mail: oboe@gest.unipd.it). tical design procedure of the time delayed control system with
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. CDOB. Firstly, this paper focuses on stability analysis of the
Digital Object Identifier 10.1109/TII.2008.2002705 control system. The effects of control parameters on stability are
1551-3203/$25.00 © 2008 IEEE
186 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

discussed using Nyquist diagrams. Characteristics of the effects


of parameters on stability come out. Then we study a practical
design procedure of the time delayed control systems. The va-
lidity of the design procedure is verified by experimental results.
Fig. 1. Time delayed control system.
In the experiment section, we also discuss the performance of
the method in the case of time-varying delay. Finally we present
a comparative study of the method to the Smith predictor.
This paper is organized as follows. In Section II, the effect of
time delay in control systems is briefly introduced. Section III
firstly describes the concept of ND. Then a time delay com-
pensation method based on the concept of ND and CDOB
is presented and the validity is demonstrated. In Section IV,
time delayed control systems used for the stability analysis
and design problem in this paper are introduced. Section V
conducts analysis about the effects of parameters in control
systems on stability and summarizes characteristics of the
effects. Then Section VI studies a practical design procedure
of the time delayed control system. Experimental results verify
the validity of the design procedure and the performance in Fig. 2. Destabilization by time delay (Nyquist diagram).
the case of time-varying delay in Section VII. Furthermore, a
comparative study of the time delay compensation method to
the Smith predictor is presented in Section VIII. Finally, this Time delay element is included in characteristic polynomial.
paper is concluded in Section IX. Therefore, the control system possibly becomes unstable. Fig. 2
demonstrates destabilization caused by time delay ( and
are the same as that in Fig. 7). It turns out that as time
II. EFFECT OF TIME DELAY delay becomes larger, the control system tends to be unstable.

In this section, the characteristic property of time delay ele- III. TIME DELAY COMPENSATION
ment in control systems and the destabilization of control sys-
In this section, we introduce a time delay compensation
tems caused by time delay are introduced.
method based on the concept of network disturbance (ND)
In the Laplace domain, time delay element is described as
and communication disturbance observer (CDOB) [23], [25].
exponential function ( is a value of delay time and means
Firstly, the concept of ND is introduced. Then estimation of
the Laplace operator)
ND by CDOB is presented and the time delay compensation
method is described. Finally, a Nyquist diagram demonstrates
the validity of the time delay compensation method.

A. Concept of ND
The characteristic property of the element in the frequency do-
main is therefore derived as follows: Fig. 3(a) shows an example of time delayed multi-input multi-
output (MIMO) system with both input delay and output
delay . The system is an extension to MIMO of the part be-
tween and in Fig. 1 (time delay is separated
(1) into input delay and output delay ). The transformations
(2) described below can be realized also in the case of single-input
single-output (SISO) systems. The state equation and the output
Equation (2) shows that the phase of time delay element delays equation are given as follows:
in proportion to both angular frequency and delay time . It
means that time delay element seriously affects phase character-
istic of control systems. The phase delay induces deterioration (4)
of the system and possibly leads to destabilization. Fig. 1 is a (5)
single-input single-output (SISO) time delayed control system
model. In Fig. 1, , and mean reference signal, Here is a state vector, is a input vector, is a
input signal, and output signal, respectively. Blocks of and output vector, is a matrix, is a matrix,
are a controller and a controlled plant. The closed-loop and is a matrix. Then the Laplace transformations are
transfer function is given as follows: described as follows:

(6)
(3)
(7)
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 187

Here, considering ND formulation (12), the state equation


changes as follows:

(15)

The Laplace transformations of the state equation (15) and the


output equation (14) are therefore described as follows:

(16)
Fig. 3. Concept of ND. (a) System with time delay. (b) System with ND.
(17)

Consequently, the transfer function of input-output relationship


Here and are described as follows ( is a is derived
unit matrix):

(8)
(18)
(9)
Since (10) is identical with (18), the system shown in Fig. 3(a) is
Consequently, the transfer function matrix of the input-
equivalent to the system shown in Fig. 3(b) from the perspective
output relationship is derived
of input-output relationship. Therefore, the concept of ND is
valid in input-output relationship. Furthermore, in the case of
only input delay or only output delay , the
concept is still correct. Consequently, we can regard the systems
(10) with time delay as the systems with ND.
From a perspective of input-output relationship, the derived B. Time Delay Compensation Method
transfer function shows that the system can be regarded as the
serial connection of and . ND is estimated by CDOB. The estimated ND is utilized
Then we reconsider the system shown in Fig. 3(a). In for time delay compensation as shown in Fig. 4. Fig. 4 shows
Fig. 3(a), taking into account only the input-output relationship, a schematic diagram of the time delay compensation method
the inside of the system can be regarded as shown in Fig. 3(b). applied to a time delayed control system shown in Fig. 1. For
There are no time delays in serial part of in Fig. 3(b). simplicity, we explain the method using an example of SISO
Time delays are included in a disturbance formulated as follows system. Detailed description of actual SISO and MIMO systems
(in both the Laplace domain and the time domain) are presented in Section IV. In Fig. 4, it is assumed that there
exists only output delay as described below for easy under-
standing ( or in Fig. 4 means a state variable of )
(11)
(12)
(19)
We define the disturbance as ND. Based on the concept of ND,
the variation of the output from the nondelayed output caused by If CDOB is ideal [i.e., the cutoff frequency of the low-pass filter
the delay can be regarded as the error caused by ND. It should be (LPF) in CDOB is assumed to be ideally infinity; see the detailed
noted that the defined ND is not independent from the system internal structure of CDOB in Section IV], the transfer function
input or the state, although disturbance is usually an external of the control system is given as follows:
signal. Here, we regard that ND is a technical term, the meaning
of which is not the same as usual external disturbance, but a
disturbance depending on the system input or the state. (20)
From here, the correctness of the concept of ND in input-
output relationship is proven. The state equation and the output Since there is no time delay element in characteristic polyno-
equation of Fig. 3(b) are given as follows: mial, time delay compensation is achieved. The effectiveness is
the same as that of the Smith predictor [13]. Furthermore, the
compensation method is able to be implemented without time
(13) delay model or time delay measurement. Therefore, the method
(14) is able to be easily implemented to various applications. The
188 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

Here, is a gravity torque, is an internal torque, is a


Coulomb friction, is a viscous friction, and is an ex-
ternal torque. Then a disturbance that is estimated by DOB
consists of load torque and parameter variations of inertia
and torque coefficient ( and mean nominal values)

(22)

Consequently, the estimated disturbance is described as fol-


Fig. 4. Schematic diagram of time delay compensation. lows:

(23)

is a cutoff frequency of LPF in DOB. Considering the feed-


back of estimated disturbance, Fig. 6(a) is equivalently trans-
formed as shown in Fig. 6(b). In Fig. 6(b), is a
high-pass filter (HPF) which shows how the disturbance affects
the system. Furthermore, it should be noted that parameters
and are nominalized in Fig. 6(b). If the value of is
assumed to be ideally infinity, Fig. 6(b) changes as shown in
Fig. 6(c). As shown in Fig. 6(c), there is no effect of distur-
bance and there are also no parameter variations. Therefore, the
Fig. 5. Validity of time delay compensation (Nyquist diagram).
acceleration reference value is realized as actual accelera-
tion of the real controlled system. Furthermore, when DOB is
comparative study between the method and the Smith predictor implemented to each joint or each controlled system (like in the
is presented in Section VIII. case of multi-DOF manipulators), it also compensates the in-
teraction of joints or systems within the bandwidth of LPF in
C. Validity of Compensation DOB. Then each system is decoupled as independent system
For the verification of the validity of the time delay compen- within the bandwidth. Actually, there are some limitations on
sation method, a Nyquist diagram is shown. In Fig. 2, the system the bandwidth or the maximal value (which can be set) of the
becomes more unstable as the value of becomes larger. Fig. 5 cutoff frequency of LPF in DOB, which are caused by the spec-
is a Nyquist diagram of the case with/without the time delay ifications of devices like the resolution of measurement equip-
compensation method ( and are the same as those in ments, processing speed of control PC, and so on. Therefore, if
Section IV). It turns out that the unstable case ( ms) in we implement appropriate devices, the specifications of which
Fig. 2 is able to be stabilized by the time delay compensation are enough for realizing sufficiently wide bandwidth or large
method. cutoff frequency of LPF in DOB for motion of the controlled
system, the decoupling within enough wide bandwidth for mo-
IV. SYSTEM DESCRIPTION tion is accomplished. It means that even if there are nonlinear
In this section, a SISO system model and a MIMO system and uncertain interactions between each system, the system is
model of time delayed control systems used in this paper are de- able to be decoupled and there is no interaction any more. It
scribed. First of all, a disturbance observer (DOB) that is locally should be noted that the time-delay compensation method by
implemented in each controlled system (or joint) is introduced CDOB needs an exact inverse model of the controlled system.
and the realization of robust acceleration control is described Therefore, in case that the system is nonlinear or has some un-
[26], [27]. Then a SISO system model and a MIMO system certainty, the method cannot be directly applied. However, DOB
model are presented in order. is able to nominalize and linearize nonlinear and uncertain sys-
tems. Then we are able to apply the time-delay compensation
A. Disturbance Observer method by CDOB to those nonlinear or uncertain systems by
Fig. 6(a) shows a schematic diagram of DOB. In Fig. 6(a), using DOB.
, and are an acceleration reference In stability analysis (Section V) and design procedure
value, a current reference value, disturbance-compensation (Section VI), for focusing on only characteristics of time de-
current, angular velocity response, and angle response, re- layed control systems, it is assumed that the cutoff frequency
spectively. Here in Fig. 6(a) is a load torque of the system of LPF in DOB is sufficiently large and robust acceleration
described as follows: control is achieved in each system. (In the experiment, we
set as 1000 rad/s that is sufficiently large.) It means that
we regard that the system is nominalized and each system is
(21) decoupled in multi-DOF system.
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 189

Since there is no time delay element in characteristic polyno-


mial, time delay is completely compensated. In this ideal case,
the closed loop transfer function is a serial connection of
(closed loop nondelayed rational transfer function) and .
Therefore, those two elements is able to be analyzed indepen-
dently. However, there inevitably exist some constraints in ac-
tually implemented systems. In other words, must be set as fi-
nite value (the bandwidth is finite). In such cases, the constraint
affects the effectiveness of the time delay compensation. The
closed loop transfer function changes to the following equation:

(29)

Since there is a time delay element in characteristic polynomial,


the possibility of destabilization by time delay arises. [It should
be noted that the characteristic polynomial is equal to (3) in the
case of (no compensation).] Therefore, the transfer func-
tion is not able to be divided into rational transfer function part
and time delay element. Then the analysis cannot be conducted
independently and the effects of parameters are not clearly per-
Fig. 6. Realization of robust acceleration control. (a) Disturbance observer. (b) ceived from the transfer function. The Nyquist method is conse-
Acceleration control system. (c) Ideal acceleration control system.
quently used in the following section for analyzing the stability
of the system diagrammatically.
B. SISO System
C. MIMO System
Fig. 7(a) shows a 1DOF inertial system (rotary manipulator)
used in this paper as SISO system. Since DOB is implemented Fig. 8(a) shows a 2DOF manipulator used in this paper as
in the system (the part described “Real System” in Fig. 7(c)), MIMO system. Since DOB is implemented in the system (the
the system is an acceleration control system shown in Fig. 7(b). part described “Real System” in Fig. 8(c)), the system is a
Then whole block diagram of the time delayed SISO control MIMO (2-input 2-output) acceleration control system shown
system is shown in Fig. 7(c) ( is a reduction gear ratio). (Since in Fig. 8(b) and each system is decoupled. The whole block
the value of the torque reference is directly input to the system, diagram of the time delayed MIMO control system is shown
there is no torque coefficient in this system.) The velocity in Fig. 8(c). Since each system (a system from to
information is calculated by pseudo derivative ( is a cutoff and a system from to ) is decoupled,
frequency) the parameter matrices in Fig. 8(c) are described as follows:

(24)
(30)
There is an LPF in CDOB ( is a cutoff frequency of LPF)
(31)

(25) (32)

If the value of is ideally infinity , the closed loop (33)


transfer function is given as follows ( is a natural angular
frequency and is a damping coefficient of the control system): Also, pseudo derivative filter matrix and LPF matrix in
CDOB are given as follows:

(26)
(34)
(27)
(35)
(28)
190 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

Fig. 8. MIMO system. (a) Picture of MIMO system (2DOF manipulator). (b)
Schematic diagram of MIMO acceleration control system. (c) Whole block di-
agram of MIMO control system.

Fig. 7. SISO system. (a) Picture of SISO system (1DOF manipulator). (b)
Schematic diagram of SISO acceleration control system. (c) Whole block eters (natural angular frequency and damping coefficient )
diagram of SISO control system. and cutoff frequency of LPF in CDOB . At first, open-loop
transfer function of the system is derived for stability analysis
using Nyquist diagram. The open-loop transfer function of SISO
Then the closed-loop transfer function matrix is de- [see Fig. 7(c)] and MIMO [see Fig. 8(c)] systems is given as fol-
scribed as follows: lows:

(36) (39)

(37)
For convenience of stability analysis focusing on , and ,
the open-loop transfer function is transformed
(38)

The transfer functions of both systems (37) and (38) are the
same as (29). Therefore, the stability analysis of MIMO system (40)
is able to be conducted by the same way as that of SISO system.
The effects of parameters ( , and ) on stability are ana-
lyzed here using Nyquist diagrams.
V. STABILITY ANALYSIS 1) Effect of Natural Angular Frequency: Fig. 9(a) is a
This section presents stability analysis of the time delayed Nyquist diagram changing natural angular frequency . Here,
SISO and MIMO control systems shown in Figs. 7(c) and 8(c). two other parameters are fixed as and . Fig. 9(a)
As described in the previous section, the results of stability anal- shows that the increase of causes decrease of gain margin
ysis in this section are valid for both SISO system and MIMO and increase of phase margin.
system. The control system has some parameters to be designed 2) Effect of Damping Coefficient: Fig. 9(b) shows a Nyquist
arbitrarily. Then we focus on two controller-associated param- diagram changing damping coefficient . In the case, two other
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 191

TABLE I
EFFECTS OF PARAMETERS ON STABILITY

Fig. 10. Design procedure in the case of T = 0 ms. (a) Nyquist diagram. (b)
Gain diagram.

VI. DESIGN PROCEDURE


From stability analysis, it turns out that three parameters (
, and ) have different effects on stability. In this section, we
focus on and study design procedure of the systems consid-
ering stability and performance. In order to make a design proce-
dure simple, controller-associated parameters (natural angular
frequency and damping coefficient ) are fixed as
and (i.e., and ). As case studies,
we deal with three cases of delays, ms (for reference),
ms, and ms. In each case, we show a Nyquist
diagram for stability design and a gain diagram for performance
design. It should be noted that the design procedure is applicable
to both SISO system (see Fig. 7) and MIMO system (see Fig. 8)
because transfer functions (29), (37), and (38) are the same.

A. Case1 ms
Fig. 10(a) shows a Nyquist diagram in the case of ms.
It turns out that has no effect on stability. Fig. 10(b) is a gain
diagram in the case of ms. From Fig. 10(b), does not
affect performance. Therefore, it is found out that has no effect
Fig. 9. Effects of parameters on stability (Nyquist diagram). (a) Effect of nat-
ural angular frequency ! . (b) Effect of damping coefficient  . (c) Effect of on both stability and performance in the case of ms. This
cutoff frequency of CDOB g . is because the value of affects the performance of time delay
compensation.

parameters are fixed as and . It is found out B. Case2 ms


that the increase of causes increase of both gain margin and Fig. 11(a) is a Nyquist diagram in the case of ms.
phase margin. The figure demonstrates that the system is near the stability limit
3) Effect of Cutoff Frequency of CDOB: Fig. 9(c) shows a when . Therefore, should be designed as follows for
Nyquist diagram changing cutoff frequency of LPF in CDOB stability assurance:
. Here, two other parameters are fixed as and .
The increase of leads to increase of gain margin and decrease
of phase margin. It should be noted that the characteristic of the (41)
effect of is opposite to that of natural angular frequency .
The three analytical results demonstrate the effects of three In the case of , the system is stable since the system has
parameters ( , and ) on stability. The results are summa- some gain margin and large phase margin. However, it turns out
rized in Table I. It turns out that the effect of has an opposite from Fig. 11(b) that the peak value of gain diagram is not suffi-
characteristic to that of . ciently suppressed when . In the case of , the peak
192 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

TABLE II
PARAMETERS OF SISO SYSTEM (1DOF MANIPULATOR)

TABLE III
Fig. 11. Design procedure in the case of T = 100 ms. (a) Nyquist diagram. PARAMETERS OF MIMO SYSTEM (2DOF MANIPULATOR)
(b) Gain diagram.

Fig. 12. Design procedure in the case of T = 500 ms. (a) Nyquist diagram. time delay in control systems. In experiments of MIMO, we use
(b) Gain diagram.
0.1 rad step command, though 1.0 rad step command is used for
SISO. The difference of the value is caused by mechanical con-
straints of MIMO system (2DOF manipulator).
value of the gain diagram is sufficiently suppressed (the perfor-
mance is improved). Those analytical results are experimentally A. Case1 ms
verified in Section VII.
Fig. 13 shows experimental results of SISO system, joint1
C. Case3 ms of MIMO system, and joint2 of MIMO system in the case of
ms. Three values of that are discussed in design proce-
Fig. 12(a) is a Nyquist diagram in the case of ms. dure are tested. In all cases, the system is stable because there
The system is close to the stability limit when . For is no time delay. However, there are steady-state errors in SISO
stability assurance, should be set as follows: system results and joint2 of MIMO system results except for
the cases of . Therefore, it turns out that the time delay
compensation method by CDOB possibly induces steady-state
(42) errors. The reason is supposed that there is limitation of cutoff
frequency of DOB . Because of the limitation, system model
The system has certain gain margin and enough phase margin in errors between the nominalized real controlled system by DOB
the case of . Then Fig. 12(b) represents a gain diagram and system model in CDOB arises. Then the model error intro-
in the case of ms. It turns out that the peak value of duces the steady-state error. Besides, especially about joint2 of
the gain diagram when is not suppressed enough. The MIMO system, it has serious mechanical loss. The point is also
peak value is sufficiently suppressed in the case of . It the reason of the steady-state error. Except for the steady-state
means that the performance is enough improved in the case of error, does not affect the system in this case.
. Experimental results verify the analytical results in
Section VII.
B. Case2 ms
Fig. 14 shows experimental results of SISO system, joint1
VII. EXPERIMENT
of MIMO system, and joint2 of MIMO system in the case of
For the verification of the design results in Section VI, three ms. Three values of studied in design procedure
cases of experiments ( ms, ms, and are tested. As the results of design procedure, all results in the
ms) are conducted. Furthermore, as a practical application, we case of are unstable. In the results of joints of MIMO
conduct an experiment in the case of time-varying delay. In system, the experiment stops at around 2.5s because of hazard
each case, the experiments of both SISO system and MIMO avoidance. In the case of , although the responses are os-
system are conducted. The parameters of experimental system cillatory, the systems is stable. Furthermore, those oscillations
are shown in Tables II and III. The controller parameter is the are sufficiently suppressed in the case of (the perfor-
same as that in previous section ( and ). mance is improved). Those results verify the validity of the de-
Time delays are simulated in PC. The signal data are temporarily sign procedure. However, the steady-state error arises especially
stored like buffering. The duration of storing is assumed to be in the result of joint2 of MIMO system. Although the reason is
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 193

Fig. 14. Experimental results (T = 100 ms). (a) SISO system. (b) MIMO
Fig. 13. Experimental results (T = 0 ms). (a) SISO system. (b) MIMO system system (joint1). (c) MIMO system (joint2).
(joint1). (c) MIMO system (joint2).

D. Application (Time-Varying Delay)


mainly the mechanical loss of the joint, we should handle the
steady-state error problem in all results. As a practical application, the experiment in the case of time-
varying delay is conducted. The time-varying delay shown as
C. Case3 ms follows is simulated in PC:

Fig. 15 demonstrates experimental results of SISO system,


joint1 of MIMO system, and joint2 of MIMO system in the case random
of ms. Three cases of the values of discussed in
design procedure are studied. As described in design procedure, For generating time-varying delay, we fluctuate the simulated
all results in the case of are almost unstable or near delay by using uniform pseudo-random number which is gen-
the stability limit. However, in the case of , although erated by mixed congruential method. Then the time-varying
there are some oscillations, all results are stable. Furthermore, delay has rectangular or uniform distribution as shown in
there are no oscillations in the case of . It means that Fig. 16. In the case of time-varying delay, there is no useful
the performance is improved as the results of design procedure design procedures of . Then, since the average of the
indicate. Also in this case, the steady-state errors are serious time-varying delay is 500 ms, we used the values of studied
especially in the results of joint2 of MIMO system. in design procedure of Case3 ms.
194 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

Fig. 16. Time-varying delay (450 ms  T : random  550 ms).

Fig. 15. Experimental results (T = 500 ms). (a) SISO system. (b) MIMO
system (joint1). (c) MIMO system (joint2).

Fig. 17 shows experimental results of SISO system, joint1


of MIMO system, and joint2 of MIMO system in the case of
the time-varying delay. Although all the results look oscilla-
tory, there are no serious oscillations. The oscillatory appear-
ance because of the time-varying delay does not affect the sta-
bility. Therefore, time-varying delay does not directly lead to
destabilization. Comparing to the results of Case3 ms,
the results are similar. Some results are almost the same except 
Fig. 17. Experimental results in the case of time-varying delay (450 ms T :
for the oscillatory appearance. As the results of design proce- random  550 ms). (a) SISO system. (b) MIMO system (joint1). (c) MIMO
system (joint2).
dure, the system is near the stability limit in the case of .
Then in the case of , although there are some oscillations
which are different from that caused by time-varying delay, the
system is stable. Furthermore, those oscillations are suppressed E. Summary
in the case of . However, also in the case of time-varying All the experimental results verify the validity of the design
delay, there is steady-state error problem. procedure. Furthermore, the design procedure is expected to be
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 195

valid also in the case of time-varying delay. However, for more


practical application to NBCSs, we should handle other dis-
tributions of time-varying delay or network delay. The differ-
ence of the distribution possibly affects the performance. Also,
the steady-state error is a serious problem. Although the main
reason of the steady-state error of joint2 of MIMO system is
supposed to be the problem of the mechanical part, the reason
of other cases may be model error. Therefore, we should handle
also the steady-state problem caused by model error. Usually,
I-controller is a good candidate for handling the steady-state
error problem. However, a simple I-controller which is directly
added to PD controller (consequently PID controller) does not
work in those cases. In those cases, CDOB unfortunately esti-
mates both ND and the disturbance caused by model error be-
tween system model in CDOB and real system in the experiment
(even if the cutoff frequency of low-pass filter in DOB is high
enough for disturbance rejection, the value is not infinity, so the
model error arises). Therefore, when the time-delay compensa-
tion is conducted using the estimated value (which includes the
disturbance caused by model error), the feedback signal also
includes errors. For example, in Fig. 13(c) (in the case of no
delay), only the curve of (no compensation) does not have
steady-state error. It means that the estimated value of CDOB
and the feedback signal which is compensated by the value have
some errors, although the estimated value by CDOB in the case
of no delay is theoretically 0. Also, the fact that the amount of
the steady-state errors is different in each case depending on
the value of indicates that the estimated disturbance caused
by model error affects the compensated feedback signal. There-
fore, even if an I-controller is directly added to the controller, Fig. 18. Time delay compensation by CDOB. (a) Time delay compensation
it would not work. However, an I-controller which is added as method by CDOB. (b) Equivalent transformation of (a). (c) Equivalent transfor-
mation of (a) in ideal situation (cutoff frequency of LPF L(s) is ideally infinity).
another controller (controller for reference and feedback signal
without compensation) like that in [24] would be one of the so-
lutions of the steady-state error problem. method by CDOB is clarified. The method by CDOB does not
need delay time model, though the method by the Smith pre-
VIII. COMPARISON TO THE SMITH PREDICTOR dictor needs delay time model. However, practically we con-
In this section, we present a comparative study of the time duct stability analysis and design with knowledge of time delay
delay compensation method by CDOB to the Smith predictor. like the previous part of this paper. It means that the method
The Smith predictor [13] is a useful time delay compensation by CDOB uses time delay model or time delay value for design
method. It compensates time delay with time delay model and of , though it does not use time delay model for implemen-
system model. On the other hand, the time delay compensation tation. The important point is that if the value of can be set
method by CDOB uses only system model. Therefore, we focus as sufficiently large value (enough margin from stability limit),
on time delay model issue in order to clarify the advantage of we do not need to consider the value of time delay. Although
the method by CDOB. For simplicity, it is assumed that there is the stability analysis and the design procedure is important in
no system model error. At first, two methods are compared the- case that is not able to be set as large enough (in the case of
oretically by using block diagrams and Nyquist diagrams. Then, implementation with cheap devices), the method does not need
some experimental results are shown to validate the comparison. delay time value in case that is able to be set as large enough
value to regard the actual system as the ideal system shown in
A. Theoretical Comparison Fig. 18(c). (The word “cheap device” means a device, the spec-
Fig. 18(a) shows a block diagram of the time delay compen- ification of which is not enough for setting as enough large,
sation method by CDOB. The block diagram is equivalently for example, low resolution encoder, low processing-speed PC,
transformed as shown in Fig. 18(b). If there is ideally no lim- and so on.) Actually, it is easy to set the value of as larger than
itation of the cutoff frequency of LPF , Fig. 18(b) 1000 rad/s in our experimental system. Then, we conducted a
can be transformed as shown in Fig. 18(c). On the other hand, stability check in the case of for the system with time
the block diagram of the time delay compensation method by delay ms, ms, and ms. (Since
the Smith predictor is Fig. 19(a) and the equivalent transforma- is enough for the case of ms, is
tion is Fig. 19(b) ( is the time delay model of the Smith pre- sufficiently large in this case.) Fig. 20(a) shows the result. The
dictor). Comparing Fig. 18(c) to Fig. 19(b), the advantage of the system is stable in all cases. On the other hand, the result of
196 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 4, NO. 3, AUGUST 2008

Fig. 19. Time delay compensation by the Smith predictor. (a) Time delay com-
pensation method by the Smith predictor. (b) Equivalent transformation of (a).

Fig. 21. Experimental results of time delay compensation methods [left:


CDOB (g = 500), right: Smith predictor (T^ = 300 ms)]. (a) SISO system.
(b) MIMO system (joint1). (c) MIMO system (joint2).

results. Although there are some steady-state errors, all results


of CDOB (left-hand side) are stable. On the other hand, in the
results of the Smith predictor (right-hand side), the system is
stable only in the case of ms in each figure and other
cases ( ms and ms: with delay time model
errors) are unstable. (In the results of joints of MIMO system,
the unstable experiments stop before 3.0 s because of hazard
avoidance.) In conclusion, the advantage of the time delay com-
pensation method by CDOB comparing to the Smith predictor
is that the method by CDOB works even if the accurate value of
time delay is unknown.

IX. CONCLUSION
This paper presented stability analysis and studied a practical
Fig. 20. Nyquist diagrams of time-delay compensation methods. (a) CDOB design procedure of time delayed control systems with CDOB.
(g = 500). (b) Smith predictor (time delay model: T^ = 300 ms).
Results of stability analysis demonstrated the characteristics of
the effects of three parameters of the control system on stability.
Then a design procedure of the control system was studied. The
stability check of the Smith predictor is shown in Fig. 20(b). experimental results verified the validity of the design proce-
In Fig. 20(b), we set the time delay model of the Smith pre- dure. In the experiment section, we also discussed the perfor-
dictor as ms. Therefore, there are ms time delay mance of the control system in the case of time-varying delay.
model errors in the cases of ms and ms. Finally, a comparative study of the time delay compensation
From Fig. 20(b), it turns out that only in the case of method by CDOB to the Smith predictor was presented.
ms (no model error) is stable and other cases (with time delay For more practical application of the method, we should also
model errors) are unstable. handle other distributions of time-varying network delay. The
difference of distributions affects the performance of the system.
B. Experimental Comparison
Therefore, we will conduct experimental studies by using real
Here we present experimental study of the comparison to val- networks as communication line in the future. Also, the steady-
idate the theoretical results. Fig. 21 [left: CDOB , state error problem of the time-delay compensation method is
right: the Smith predictor ( ms)] shows experimental also important for further applications.
NATORI et al.: DESIGN OF TIME DELAYED CONTROL SYSTEMS WITH CDOB 197

REFERENCES [22] T. Ishihara and H. J. Guo, “LTR design of integral controllers for time-
delay plants using disturbance cancellation,” in Proc. 44th Conf. De-
[1] J. Baillieul and P. J. Antsaklis, “Control and communication challenges cision and Control, and the Eur. Control Conf., Seville, Spain, Dec.
in networked real-time systems,” Proc. IEEE, vol. 95, no. 1, pp. 9–28, 2005, pp. 1683–1688.
Jan. 2007. [23] K. Natori, T. Tsuji, K. Ohnishi, A. Hace, and K. Jezernik, “Robust bi-
[2] W. Zhang, M. S. Branicky, and S. M. Phillips, “Stability of networked lateral control with internet communication,” in Proc. 30th Annu. Conf.
control systems,” IEEE Control Syst. Mag., vol. 21, no. 1, pp. 84–99, IEEE Ind. Electron. Soc., Busan, Korea, Nov. 2004, pp. 2321–2326.
Feb. 2001. [24] K. Natori and K. Ohnishi, “An approach to design of feedback systems
[3] S. Soucek and T. Sauter, “Quality of service concerns in ip-based con- with time delay,” in Proc. 31st Annu. Conf. IEEE Ind. Electron. Soc.,
trol systems,” IEEE Trans. Ind. Electron., vol. 51, no. 6, pp. 1249–1258, Raleigh, NC, Nov. 2005, pp. 1931–1936.
Dec. 2004. [25] K. Natori, R. Oboe, and K. Ohnishi, “Analysis and design of time de-
[4] K. C. Lee, S. Lee, and M. H. Lee, “QoS-based remote control of net- layed control systems with communication disturbance observer,” in
worked control systems via profibus token passing protocol,” IEEE Proc. IEEE Int. Symp. Industrial Electronics, Vigo, Spain, Jun. 2007,
Trans. Ind. Informat., vol. 1, no. 3, pp. 183–191, Aug. 2005. pp. 3132–3137.
[5] Y. Tipsuwan and M.-Y. Chow, “Gain scheduler middleware: A [26] K. Ohnishi, M. Shibata, and T. Murakami, “Motion control for ad-
methodology to enable existing controllers for networked control and vanced mechatronics,” IEEE/ASME Trans. Mechatron., vol. 1, no. 1,
teleoperation—Part I: Networked control,” IEEE Trans. Ind. Electron., pp. 56–67, Mar. 1996.
vol. 51, no. 6, pp. 1218–1227, Dec. 2004. [27] T. Murakami, F. Yu, and K. Ohnishi, “Torque sensorless control in
[6] Y. Tipsuwan and M.-Y. Chow, “Gain scheduler middleware: A multidegree-of-freedom manipulator,” IEEE Trans. Ind. Electron., vol.
methodology to enable existing controllers for networked control and 40, no. 2, pp. 259–265, Apr. 1993.
teleoperation—Part II: Teleoperation,” IEEE Trans. Ind. Electron.,
vol. 51, no. 6, pp. 1228–1237, Dec. 2004. Kenji Natori (S’06) received the B.E. degree in
[7] D. Yue, Q.-L. Han, and J. Lam, “Network-based robust H control of system design engineering and the M.E. degree in
systems with uncertainty,” Automatica, vol. 41, no. 6, pp. 999–1007, integrated design engineering from Keio University,
2005. Yokohama, Japan, in 2004 and 2006, respectively,
[8] L. S. Hu, T. Bai, P. Shi, and Z. Wu, “Sampled-data control of networked where he is currently pursuing the Ph.D. degree in
linear control systems,” Automatica, vol. 43, no. 5, pp. 903–911, 2007. integrated design engineering.
[9] C.-L. Hwang, L.-J. Chang, and Y.-S. Yu, “Network-based fuzzy decen- Since 2007, he has been a Research Fellow of the
tralized sliding-mode control for car-like mobile robots,” IEEE Trans. Japan Society for the Promotion of Science. His re-
Ind. Electron., vol. 54, no. 1, pp. 574–585, Feb. 2007. search interests include time delay systems, network-
[10] W. Hu, G. Liu, and D. Rees, “Event-driven networked predictive con- based control systems (NBCSs), and haptics.
trol,” IEEE Trans. Ind. Electron., vol. 54, no. 3, pp. 1603–1613, Jun.
2007.
[11] P. Martí, J. Yépez, M. Velasco, R. Villà, and J. M. Fuertes, “Managing
quality-of-control in network-based control systems by controller and
message scheduling co-design,” IEEE Trans. Ind. Electron., vol. 51, Roberto Oboe (M’04–SM’07) received the Laurea
no. 6, pp. 1159–1167, Dec. 2004. degree and the Doctorate degree in electrical engi-
[12] J. P. Richard, “Time-delay systems: An overview of some recent neering from the University of Padova, Padova, Italy
advances and open problems,” Automatica, vol. 39, no. 10, pp. in 1988 and 1992, respectively.
1667–1694, 2003. From 1993 to 2003, he was with the Dipartimento
[13] O. J. M. Smith, “A controller to overcome dead time,” ISA J., vol. 6, di Elettronica e Informatica of the University of
no. 2, pp. 28–33, 1959. Padova. He is currently an Associate Professor at
[14] K. Watanabe and M. Ito, “A process-model control for linear systems the Dipartimento di Tecnica e Gestione dei Sistemi
with delay,” IEEE Trans. Autom. Control, vol. AC-26, no. 6, pp. Industriali, Universita di Padova, Vicenza, Italy. His
1261–1269, Dec. 1981. interests are in the fields of data storage, parametric
[15] M. R. Stojić, M. S. Matijević, and L. S. Draganović, “A robust smith identification of mechanical systems, control and
predictor modified by internal models for integrating process with dead applications of MEMS devices, applied digital control, telerobotics, virtual
time,” IEEE Trans. Autom. Control, vol. 46, no. 8, pp. 1293–1298, Aug. mechanism, haptic devices, biomedical equipment, and applications of digital
2001. signal processors to real-time control.
[16] T. Liu, Y. Z. Cai, D. Y. Gu, and W. D. Zhang, “New modified smith
predictor scheme for integrating and unstable processes with time
delay,” Proc. Inst. Elect. Eng., Control Theory Appl., vol. 152, no. 2,
pp. 238–246, Mar. 2005. Kouhei Ohnishi (S’78–M’80–SM’00–F’01) re-
[17] W. Zhang and C. Lin, “Multivariable smith predictors design for non- ceived the B.E., M.E., and Ph.D. degrees in electrical
square plants,” IEEE Trans. Control Syst. Technol., vol. 14, no. 6, pp. engineering from the University of Tokyo, Tokyo,
1145–1149, Nov. 2006. Japan, in 1975, 1977, and 1980, respectively.
[18] G. J. Silva, A. Datta, and S. P. Bhattacharyya, “On the stability and Since 1980, he has been with the Department of
controller robustness of some popular PID tuning rules,” IEEE Trans. System Design Engineering, Keio University, Yoko-
Autom. Control, vol. 48, no. 9, pp. 1638–1641, Sep. 2003. hama, Japan. His research interests include motion
[19] A. N. Gündeş, H. Özbay, and A. B. Özgüler, “PID controller synthesis control, mechatronics, and robotics.
for a class of unstable MIMO plants with I/O delays,” Automatica, vol. Prof. Ohnishi is the recipient of the Best Paper
43, no. 1, pp. 135–142, 2007. Awards from the Institute of Electrical Engineers
[20] E. Fridman and U. Shaked, “A descriptor system approach to H con- of Japan and the Japan Society for Precision
trol of linear time-delay systems,” IEEE Trans. Autom. Control, vol. 47, Engineering, and the Outstanding Paper Awards from the IEEE Industrial
no. 2, pp. 253–270, Feb. 2002. Electronics Conferences (IECON.85, IECON.92, and IECON.93). He is also
[21] S. Xu, J. Lam, and Y. Zou, “New results on delay-dependent robust the recipient of the EPE-PEMC Council Award and the Dr.-Ing. Eugene
H control for systems with time-varying delays,” Automatica, vol. Mittelmann Achievement Award from the IEEE Industrial Electronics Society
42, no. 2, pp. 343–348, 2006. in 2004.

You might also like