You are on page 1of 10

380 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO.

1, JANUARY 2008

Improving Disturbance-Rejection Performance Based


on an Equivalent-Input-Disturbance Approach
Jin-Hua She, Member, IEEE, Mingxing Fang, Yasuhiro Ohyama, Member, IEEE,
Hiroshi Hashimoto, Member, IEEE, and Min Wu

Abstract—This paper presents a new method of improving the input channel. However, the existence of an EID has never been
disturbance-rejection performance of a servo system based on discussed from a theoretical standpoint, and the estimation of an
the estimation of an equivalent input disturbance (EID). First, the EID is meaningful only when its existence is guaranteed. In this
concept of EID is defined. Next, the configuration of an improved
servo system employing the new disturbance-estimation method is paper, a lemma is first presented that theoretically guarantees
described. Then, a method of designing a control law employing the existence of an EID for a plant under some assumptions.
a disturbance estimate is explained. Finally, the speed control of a Next, a new method of estimating an EID is explained that
rotational control system is used to demonstrate the validity of the overcomes the drawbacks of, and has several advantages over,
method, and some design guidelines are presented. existing ones. For example, it can reject any kind of distur-
Index Terms—Disturbance estimation, disturbance rejection, bance; it does not require the differentiation of measured
equivalent input disturbance (EID), perfect regulation, servo outputs, rank conditions, or priori information on a disturbance;
system, stable inversion. and it does not use the inverse dynamics of the plant, thereby
avoiding the cancellation of unstable poles/zeros. The method
I. I NTRODUCTION only needs the input and output of the plant to produce an EID
and does not require the state of the plant. This enlarges its
O VER THE PAST few decades, a considerable number of
studies have been devoted to the estimation and rejection
of an unknown disturbance (e.g., [1]–[11]). While most of these
range of application. The system configuration is also very sim-
ple, and the Separation Theorem can be used to independently
methods require the differentiation of measured outputs, the design the state feedback, and the observer and low-pass filter.
methods in [1]–[8] do not. However, in [1]–[3], rank conditions This paper first defines an EID for a system containing
are imposed on the unknown inputs; [4] requires exact infor- disturbances that may not necessarily be imposed on the control
mation on a disturbance; [5] and [6] consider only a special input channel. Next, a new method of EID estimation based on
class of disturbances; and [7] and [8] use the inverse dynam- the control input and the output of the plant is described. Then,
ics of the plant directly in the construction of the estimator. an improved servo system employing an estimate of an EID
She et al. [12] devised a new method that overcomes the is constructed, and a design algorithm for the control system
drawbacks of these methods. However, the state of the plant is presented. Finally, the results of simulations and experiments
is needed for the estimation. on the speed control of a rotational control system are presented
On the other hand, from the standpoint of the control system, to demonstrate the validity of the method, and some design
it is more reasonable to estimate a disturbance on the control guidelines are given.
input channel than to estimate the disturbance itself because In this paper, to make block diagrams easier to understand,
we have to use the control input to improve the disturbance- only the minus inputs of a summer are marked, and the plus
rejection performance. In fact, many disturbance-rejection symbols are omitted.
methods, such as the disturbance observer, implicitly assume
that an equivalent input disturbance (EID) exists on the control
II. C ONSTRUCTION OF I MPROVED S ERVO S YSTEM

Manuscript received December 6, 2006; revised July 2, 2007. The work of This section first defines an EID and, then, describes the
J.-H. She, Y. Ohyama, and H. Hashimoto was supported in part by the Electro- configuration of an improved servo system constructed by
Mechanic Technology Advancing Foundation, Japan, by the Grant-in-Aid for inserting a disturbance estimator that estimates an EID into a
Scientific Research (C), Japan Society for the Promotion of Science (JSPS)
under Grant 18560259, and by the High Tech-Research Center Project funded conventional servo system.
by the Ministry of Education, Culture, Sports, Science and Technology of
Japan. The work of M. Fang and M. Wu was supported by the National Science
Fund for Distinguished Youth Scholars of China under Grant 60425310. A. Definition of EID
J.-H. She, Y. Ohyama and H. Hashimoto are with the School of Bionics,
Tokyo University of Technology, Tokyo 192-0982, Japan (e-mail: Consider the linear time-invariant plant in Fig. 1
she@bs.teu.ac.jp; ohyama@bs.teu.ac.jp; hiro-hashimoto@ieee.org).
M. Fang is with the School of Information Science and Engineering, Central 
South University, Changsha 410083, China, and also with the College of ẋo (t) = Axo (t) + Bu(t) + Bd d(t)
(1)
Physics and Electronic Information, Anhui Normal University, Wuhu 241000, yo (t) = Cxo (t)
China (e-mail: mingxing.fang@gmail.com).
M. Wu is with the School of Information Science and Engineering, Central
South University, Changsha 410083, China (e-mail: min@csu.edu.cn). where A ∈ Rn×n , B ∈ Rn×nu , Bd ∈ Rn×nd , C ∈ Rny ×n ,
Digital Object Identifier 10.1109/TIE.2007.905976 xo (t) ∈ Rn , u(t) ∈ Rnu , d(t) ∈ Rnd , and yo (t) ∈ Rny .

0278-0046/$25.00 © 2008 IEEE


SHE et al.: IMPROVING DISTURBANCE-REJECTION PERFORMANCE BASED ON AN EID APPROACH 381

Remark 3: Note that, since the stable-inversion approach


computes a particular control input and a particular state trajec-
tory that either does not excite the zeros of the system or excite
them in a stable manner, this approach produces a satisfactory
control input for a given output, even for a system with unstable
zeros/poles.
Fig. 1. Plant. Remark 4: Since we focus on the effect of disturbances
on the output rather than on the disturbances themselves, the
stable-inversion approach guarantees the existence of an EID,
regardless of any difference in dimension between the control
input matrix B and the disturbance input matrix Bd or any
difference between the ranges of their linear transforms.
From the aforementioned lemma, we know that we can
Fig. 2. Plant with EID. calculate an EID from the input and output of the plant. Since
the calculation is complicated and also requires information on
The following assumptions are made about the plant. future outputs, this approach cannot be directly used to obtain
Assumption 1: (A, B, C) is controllable and observable. an EID. However, since this lemma clarifies that an EID exists,
Assumption 2: (A, B, C) has no zeros on the imaginary axis. it gives us a theoretical guarantee of the meaningfulness of EID
Remark 1: Assumption 1 is for simplicity. Since we want estimation. In the next section, we devise a simple method of
to use a control input to improve the tracking performance of estimating an EID.
the output, if the plant is not controllable and/or observable, Considering the aforementioned lemma, we mainly use (2)
then we only need to consider a controllable and observable as the model of the plant in the rest of this paper.
subsystem of it.
Remark 2: Assumption 2 is necessary to guarantee the in-
ternal stability of a servo system and to allow the output of the B. Estimation of EID
plant to track a reference input without a steady-state error [13]. The configuration of an improved servo system is shown in
For simplicity, we consider the single-input single-output Fig. 3, where
(SISO) case, which means that nu = 1 and ny = 1. Note that,
since B and Bd may have different dimensions, the disturbance BT
B + := . (4)
may be imposed on a channel other than the control input BTB
channel, and the number of disturbances and associated input
channels may also be larger than one. However, if we assume It can be viewed as a conventional servo system (internal model,
that a disturbance is imposed only on the control input channel, state observer, and state feedback) combined with a disturbance
as shown in Fig. 2, then the plant is given by estimator that produces an estimate of an EID. KR and KP
are the state-feedback gains, L is the observer gain, and F (s)

ẋ(t) = Ax(t) + B [u(t) + de (t)] is a low-pass filter that limits the angular-frequency band of
(2) the disturbance estimate. Since we know the exact form of the
y(t) = Cx(t).
reference input, an internal model is employed in the control
An EID is defined as follows. system to improve the tracking precision for the reference input.
Definition 1: Let the control input be u(t) = 0. Then, the On the other hand, since the disturbance is assumed to be
output of the plant (1) for the disturbance d(t) is yo (t), and unknown, no internal model of it is contained in the control
the output of the plant (2) for the disturbance de (t) is y(t). The system, and we need to devise a method of suppressing its
disturbance de (t) is called an EID of the disturbance d(t) if influence on the output. Note that the EID estimator makes
y(t) ≡ yo (t) for all t ≥ 0. use of the state observer instead of the inverse dynamics of
Let the plant. This is a big difference in system structure between
the disturbance-observer method and the one presented in this
Φ = {pi (t) sin(ωi t + φi )} , i = 0, . . . , n, n<∞ paper.
(3) In Fig. 3, for the state observer
where ωi (≥ 0) and φi are constants, and pi (t) denotes any ˙
x̂(t) = Ax̂(t) + Buf (t) + L [y(t) − C x̂(t)] (5)
polynomials in time t (i = 0, . . . , n). If the trajectory of the
output caused by the disturbance d(t) is yo (t) ∈ Φ, then, from holds. Letting
the concept of stable inversion [14], it follows that there exists
an EID de (t) ∈ Φ on the control input channel that produces ∆x(t) = x̂(t) − x(t) (6)
the same trajectory. This leads to the following lemma.
Lemma 1: Under Assumptions 1 and 2, there always exists and substituting it into (2) yields
an EID de (t) ∈ Φ on the control input channel of the distur-
bance d(t) imposed on the plant (1), and the output it produces ˙
x̂(t) = Ax̂(t) + Bu(t) + {Bde (t) + [∆ẋ(t) − A∆x(t)]} .
belongs to Φ. (7)
382 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 1, JANUARY 2008

Fig. 3. Configuration of the improved servo system.

Assume that there exists a control input ∆d(t) that satisfies includes the effects of the disturbance. This difference degrades
the tracking precision if we directly use the available states. In
∆ẋ(t) − A∆x(t) = B∆d(t). (8) order to obtain an EID with a high precision, it is important to
guarantee that ŷ(t) − y(t) converges to zero. Therefore, using
Substituting (8) into (7) and letting the estimate of the EID be a full-order state observer is desirable.
ˆ = de (t) + ∆d(t)
d(t) (9)
C. Disturbance Rejection
allows us to express the plant as
  Combining the disturbance estimate (13) with the original
˙
x̂(t) ˆ
= Ax̂(t) + B u(t) + d(t) . (10) servo control law yields the following control law:
˜
u(t) = uf (t) − d(t) (14)
Remark 5: Equations (9) and (10) mean that, if we take the
state of the plant with an EID to be x̂(t), which is exactly the
as shown in Fig. 3. This modified control law improves the
state of the observer, then the difference between the state of
disturbance-rejection performance. The method described in
the plant and that of the observer is equivalent to the difference
this paper has two important features that are not provided by
between the exact value and the estimate of the EID. Equation
other methods.
(10) plays a key role in the EID estimation.
Equations (5) and (10) yield 1) The configuration of the system is very simple.
2) Incorporating the disturbance estimate directly into the
 
ˆ + u(t) − uf (t) = LC [x(t) − x̂(t)] .
B d(t) (11) designed servo control law makes the disturbance-
rejection performance easy to improve.
ˆ then a least square solution is Regarding the first feature, the improved servo system can
If we solve (11) for d(t),
be viewed as a conventional servo system enhanced by the
ˆ = B + LC [x(t) − x̂(t)] + uf (t) − u(t).
d(t) (12) plugging-in of a disturbance estimate. Therefore, the structure
is very simple and very easy to understand. Regarding the
ˆ is filtered by F (s), which selects the angular-frequency
d(t) second feature, let ωr be the highest angular frequency for dis-
band for disturbance estimation. Thus, the filtered disturbance turbance rejection and Ωr := {ω, 0 ≤ ω ≤ ωr } be the angular-
˜ is given by
estimate d(t) frequency band for disturbance rejection. A suitable design
ˆ converges to de (t), and
of the observer guarantees that d(t)
D̃(s) = F (s)D̂(s) (13) D̃(jω) ≈ D̂(jω) for all ω ∈ Ωr is true for a properly designed
˜ is a good approximation
low-pass filter F (s). Therefore, d(t)
where D̃(s) and D̂(s) are the Laplace transforms of d(t) ˜ and of de (t).
ˆ respectively.
d(t), Remark 7: Umeno et al. [15] presented a Q-filter method
Remark 6: Since an estimate is obtained for an EID, the for the design of a disturbance observer. The cutoff angular
channel on which the EID is imposed might be different from frequency of a Q-filter must be sufficiently higher than the
that of the actual disturbance. Therefore, generally speaking, highest angular frequency of the disturbances to guarantee that
a full-order state observer must be used to estimate the state the disturbance estimate is imposed on the control input. The
of the plant because, if a disturbance exists, then the estimated situation is the same for the low-pass filter F (s) in this paper.
state of the plant might be different from the actual state, which However, as pointed out in [16], for a Q-filter, improving the
SHE et al.: IMPROVING DISTURBANCE-REJECTION PERFORMANCE BASED ON AN EID APPROACH 383

Fig. 4. Block diagram for the design of low-pass filter and state observer.

disturbance-rejection performance requires an increase on the thus, do not appear in the transfer function (19). For this system,
order of the filter; basically, to guarantee causality, the order we obtain the following from the small-gain theorem [17].
cannot be lower than the relative degree of the plant. In con- Theorem 1: For a suitably designed state-feedback gain
trast, even a first-order F (s) produces satisfactory disturbance- [KP KR ], the control law (14) guarantees the stability of the
rejection performance. control system under the following conditions.
This system also has another very important feature: The 1) A − LC is stable.
state-feedback control law can be designed independent of the 2) GF ∞ < 1 where GF ∞ := sup0 ≤ ω ≤ ∞ σmax
observer and the low-pass filter, as long as stability is the only [G(jω)F (jω)], and σmax (G) means the maximum
concern. This is discussed in the following section. singular value of G.
Remark 8: The stability conditions for the improved servo
D. Design of Filter and State Observer system can be broken down into two parts (Separation Theorem
[18]). First, the state-feedback servo system is stable. Second,
The state observer gain L and the low-pass filter F (s) should the conditions in Theorem 1 hold. Since the only parameters
be designed so that they do not destroy the stability of the in those conditions are L and F (s), their design is much sim-
system. Regarding stability, if we let pler than the design of parameters in the disturbance-observer
r(t) = 0, d(t) = 0 (15) method [7], which requires a low-pass filter to guarantee the
stability of the whole system.
then the plant (2) is On the other hand, for the dual system of the plant

ẋ(t) = Ax(t) + Bu(t). (16) ẋL (t) = AT xL (t) + C T uL (t)
(20)
yL (t) = B T xL (t)
Combining (5), (6), and (14) with the aforementioned equation
yields consider the state feedback parameterized by a scalar ρ > 0
˜
∆ẋ(t) = (A − LC)∆x(t) + B d(t). (17) uL (t) = LT
ρ xL (t).

On the other hand, (12) is equivalent to If (AT , C T , B T ) [and, thus, (A, B, C)] is a minimum-phase
ˆ = −B + LC∆x(t) + d(t).
˜ system, then, based on the concept of perfect regulation [19],
d(t) (18)
[20], we can obtain an LT ρ that ensures
Using the relationships (15)–(18) and redrawing Fig. 3 yield
Fig. 4. Equations (17) and (18) yield the transfer function from lim [sI − (A − Lρ C)]−1 B = 0.
ρ→∞
˜ to d(t)
d(t) ˆ
Note that [sI − (A − Lρ C)]−1 B is part of G(s), which means
G(s) = 1 − B + LC [sI − (A − LC)]−1 B that a large enough ρ makes |G(jω)| sufficiently small for
= B + (sI − A) [sI − (A − LC)]−1 B. (19) all ω ∈ [0, ∞). Therefore, based on the concept of perfect
regulation, for a given F (s), we can obtain an L that satisfies
Note that the stabilized dynamics under the dotted line in Fig. 4 the conditions in Theorem 1. The design procedure is explained
are unobservable from the dynamics above the dotted line and, in the next section.
384 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 1, JANUARY 2008

E. Design Procedure
Summarizing the aforementioned results yields the following
design algorithm for the improved servo system.

Design algorithm:
Step 1) Design the feedback gains KP and KR for a conven-
tional servo system using an existing method (e.g.,
the optimal control method).
Step 2) Choose an angular-frequency band Ωr for distur-
bance rejection.
Step 3) Choose a low-pass filter that satisfies

F (jω) ≈ 1 ∀ω ∈ Ωr . (21)

Step 4) Select a ρ and calculate an Lρ that stabilizes


A − Lρ C, and then, plot |1/G(jω)| and |F (jω)|.
Step 5) Check if the second condition in Theorem 1, i.e.,

|1/G(jω)| > |F (jω)| , ω ∈ [0, ∞) (22)

is true from the Bode plots. If it is not, then increase ρ until (22)
is satisfied.
This design algorithm is intuitive and easy to understand.
Furthermore, since only one parameter ρ needs to be changed,
adjustment of the design is very simple.

III. S IMULATIONS AND E XPERIMENTAL V ERIFICATION Fig. 5. Rotational control system. (a) Model. (b) Block diagram.

We employed the method previously described for the speed In (23)–(25), the variables and parameters with the subscript p
control of a rotational control system (Fig. 5) consisting of two (or d) indicate that they are related to the controlled motor (or
dc motors: One was used as the controlled object, and the other disturbance generator). They are defined as follows:
was used as a disturbance generator. The system is very similar u Applied voltage (in volts).
to the well-known two-mass system [21]. However, in this d Disturbance voltage (in volts).
paper, we used one motor to generate a torque disturbance that ip (id ) Armature current (in amperes).
was imposed on the controlled motor from a channel other than τp (τd ) Torque produced (in newton meters).
the control input channel. The axles of the motors were coupled τpd Twisting torque (in newton meters).
together with a spring: 1) The controlled motor maintained the ωp (ωd ) Rotational speed (in radians per second).
specified rotational speed, in spite of any disturbances; and θp (θd ) Rotation angle (in radians).
2) the disturbance generator produced disturbances, adding ir- Jp (Jd ) Inertia (in kilogram square meters).
regularities to the rotational speed. The models of the controlled Rp (Rd ) Resistance of armature coil (in ohms).
motor and the disturbance generator are Kp (Kd ) Back-electromotive-force constant (in volt seconds
 per radian) or torque constant (in newton meters per

 Rp ip (t) + Kp ωp (t) = u(t) ampere).
τp (t) = Kp ip (t) (23) Kpd Twisting elasticity coefficient of coupling (in new-


Jp ω̇p (t) = τp (t) − τpd (t) ton meters per radian).
Choosing the state to be xo (t) = [ωp (t) ωd (t) θp (t) −
and θd (t)]T and the output to be yo (t) = ωp (t) yields the following
state space description in (1):

 Rd id (t) + Kd ωd (t) = d(t)   K2 K

τd (t) = Kd id (t) (24) 
 − Jp Rp p 0 − Jpd
 J ω̇ (t) = τ (t) + τ 
  p


 A= 0 K2 Kpd 
d d d pd(t) 
 − Jd Rd d Jd
1 −1 0 (26)
  Kp     T
respectively, and 
 0 1

 J R

 B =  0  Bd =  Kd  C =  0  .
p p

 Jd Rd
τpd (t) = Kpd [θp (t) − θd (t)] . (25) 0 0 0
SHE et al.: IMPROVING DISTURBANCE-REJECTION PERFORMANCE BASED ON AN EID APPROACH 385

For the experimental system, the parameters were iden-


tified to be
  

 −31.31 0 −2.833 × 104
A =  0 −10.25 8001 
(27)

 1 −1 0

B = [ 28.06 0 0 ]T , Bd = [ 0 7.210 0 ]T .

Simple verification shows that the plant is controllable and


observable, and has no zeros on the imaginary axis. The fact
that this plant is a minimum-phase system allows us to use the
aforementioned method to design an observer gain and a low- Fig. 6. Tuning of G(s) and selection of F (s).
pass filter. We let the reference input be the step signal
where ωc is the cutoff angular frequency of the filter. Then,
r(t) = 1000 × 1(t) rpm. (28) an optimal observer gain L was designed that minimized the
performance index
Therefore, the internal model in the control system is
chosen to be ∞
 
JL = ρxT 2
L (t)QL xL (t) + RL uL (t) dt
AR = 0, BR = 1.
0

The following disturbance is added to the system: QL = diag{ 1 10−6 10−6 }, RL = 1


0, t < 10 or t > 30 for the system (20) [the parameters are given in (27)]. ρ was
d(t) = 0.25×[sin 4πt + cos 2πt adjusted so that GF ∞ < 1. The resulting ρ was 106 , which
+ sin πt + sin(0.5πt)]−1, 10 ≤ t ≤ 30. yielded
(29)
L = [ 972.313 −11.57900 −0.11079 ]T .
An improved servo system was designed by following the
design procedure in the previous section. First, the disturbance It is clear from Fig. 6 that the second condition in Theorem
was ignored, and a single augmented-state representation con- 1 is satisfied, which means that the improved servo system is
taining the plant and an internal model of the step signal [22] stable.
was constructed
      
δ ẋ(t) A 0 δx(t) B A. Simulations
= + δu(t) (30)
δ ẋR (t) −BR C AR δxR (t) 0
Some simulation results are shown in Figs. 7 and 8. In
where Fig. 7, the step reference input was imposed at t = 0 s. After
the system entered the steady state, the disturbance (29) was
δx(t) = x(t) − x(+∞) imposed on the system during the time t = 10–30 s.
δxR (t) = xR (t) − xR (+∞) Since the servo system contains only a step-type internal
model, it can only reject the dc component of the distur-
δu(t) = u(t) − u(+∞). bance and merely suppresses the other components in (29) to
some extent but cannot reject them. Thus, a large steady-state
Minimizing the performance index tracking error arises (peak-to-peak value: 147.67 rpm). The
∞     corresponding control input is also shown in the same figure.
  δx(t) Our new method yields a big improvement. The simulation
JK = δxT (t)δxT
R (t) QK + RK δu2 (t) dt
δxR (t) results obtained with disturbance estimation in Fig. 8 show that
0
the disturbance is satisfactorily rejected in both the transient and
QK = diag{ 1 1 1 10 }, RK = 1 steady states. The steady-state tracking error drops to 9.15 rpm
(peak to peak), which is about 6% of that without disturbance
yields estimation. The estimated EID is also shown in the same figure.
It adjusts the control input, thereby suppressing the effects of
[KP |KR ] = [ 0.42217 0.13939 −1.47170| −3.16228 ] .
the disturbance.
Next, since the highest angular frequency contained in the
disturbance d(t) is 4π rad/s, we selected Ωr by setting ωr to B. Relationship Between Characteristics of Filter and
13 rad/s. Then, we chose a first-order filter and set T = Disturbance-Rejection Performance
0.01 s(< 1/(5ωr )). This yielded the low-pass filter
In Fig. 3, the low-pass filter F (s) is used to select the
1 1 angular-frequency band for disturbance rejection. It markedly
F (s) = , ωc = (31)
Ts + 1 T affects the disturbance-rejection performance. Here, we
386 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 1, JANUARY 2008

Fig. 7. Simulation results without disturbance estimation. Fig. 8. Simulation results with disturbance estimation.

investigate the relationship between that performance and (a)


the cutoff angular frequency and (b) the order of the filter for
the experimental system. However, the results are generally
applicable and can be used as a guideline in choosing a suitable
low-pass filter.
In the previous section, the cutoff angular frequency of the
low-pass filter was 100 rad/s. The highest angular frequency
of the disturbance in (29) was 4π rad/s. At this frequency,
|F (j4π)| = 0.99220 and ∠F (j4π) = 7.16246◦ . While the fil-
ter attenuated the amplitude of the component of 4π rad/s in
the EID very little, it slightly shifted the phase. As a result,
the disturbance-estimation method reduced the power of this
component to about 2%. Lower frequency components pass the
filter with virtually no gain attenuation or phase lag. Thus, this
filtered EID contributes a great deal to the disturbance rejection
when it is incorporated into the control law (14).
To investigate the relationship between the cutoff angular
frequency and the disturbance-rejection performance, under the
constraint that the stability of the control system is guaranteed Fig. 9. Bode plot of F (s) [(31), ωc = 100, 63, 32, 16, 8, and 2 rad/s].
(Fig. 9), we chose cutoff angular frequencies of 63, 32, 16,
8, and 2 rad/s and carried out control. Table I shows the
power spectral density of the steady-state output at the angular the component is suppressed to less than 5% of that without
frequencies of the disturbance components. Clearly, when the disturbance estimation. Thus, it is desirable to choose the cutoff
cutoff angular frequency of the low-pass filter is five times angular frequency to be five times higher than the highest
higher than a disturbance component, the component of the angular frequency of the disturbance to be rejected.
disturbance estimate passes the filter with virtually no gain In (31), a first-order filter was used to limit the angular-
attenuation or phase lag, and the power spectral density of frequency band of disturbance estimation. However, generally
SHE et al.: IMPROVING DISTURBANCE-REJECTION PERFORMANCE BASED ON AN EID APPROACH 387

TABLE I
POWER SPECTRAL DENSITY OF STEADY-STATE OUTPUT FOR VARIOUS
CUTOFF ANGULAR FREQUENCIES OF LOW-PASS FILTER (31)
(NDE: NO DISTURBANCE ESTIMATION)

speaking, it is possible to use a high-order low-pass filter,


such as
bn−1 sn−1 + · · · + b1 s + b0
F (s) = . (32)
sn + an−1 sn−1 + · · · + a1 s + a0

We investigated the relationship between the order of the low-


pass filter and the disturbance-rejection performance. Specifi-
cally, we examined the widely used Butterworth, Chebyshev, Fig. 10. Bode plot of Butterworth filter [(32), n = 1, 2, 3, 4].
Bessel, and Elliptic (Cauer) filters with orders of two, three,
TABLE II
and four [23]. Note that, when n = 1, these filters all reduce POWER SPECTRAL DENSITY OF STEADY-STATE OUTPUT FOR
to the same form (31). Since the results for these filters exhibit BUTTERWORTH FILTER (32) (NDE: NO DISTURBANCE ESTIMATION)
the same trend, we present only those for the Butterworth filter.
Fig. 10 shows a Bode plot of the filter, and Table II shows the
control results.
It is clear from Fig. 10 that, for a given cutoff frequency
(100 rad/s), the higher the order of the filter is, the lower the
angular frequency is at which phase lag occurs. Therefore,
even though the disturbance estimate passes the low-pass filter
without gain attenuation, the phase shifts, which degrades the
The experimental results with and without disturbance esti-
disturbance-rejection performance.
mation are shown in Fig. 12. A comparison of the experimental
In this paper, we require that the disturbance estimates pass
and simulation results reveals that, while both are very similar
through the filter without any damping or phase lag in the
for the conventional servo system (i.e., the control system
angular-frequency band Ωr and are not concerned about the
without disturbance estimation), the rise time was longer in
rolloff speed of the gain of F (jω) outside that band. To achieve
the experiments than in the simulations when the EID estimate
this, a first-order low-pass filter is the best. Therefore, regarding
was incorporated into the system. This can be explained as
the choice of low-pass filter, we can conclude as follows.
follows: The nonlinearities (static friction, dead zones, etc.) in
1) The structure of the low-pass filter should be F (s) = the experimental system make the initial output of the actual
1/(T s + 1). plant slower than that of the observer because the observer
2) The time constant of the filter T should be chosen to does not take the nonlinearities into account. The EID esti-
satisfy the condition T ≤ 1/(5ωr ). mator took this difference to be a disturbance, produced an
EID, and added it to the control input channel. Regarding the
experimental results, when the disturbance was removed from
C. Experimental Results
the system (t = 30–40 s), the standard deviation of the tracking
Experiments were carried out to demonstrate the validity of error was 13.66. However, during the period t = 20–30 s, the
the new method using the setup in Fig. 11. The ratings of the standard deviation of the tracking error was 57.75 without
two motors are the following: 80 W, dc 24 V, and 2500 rpm. The disturbance estimation and 22.33 with disturbance estimation.
controller was a desktop computer, and the control program was The power spectral density reveals that the components of the
written in C language and run under MS-DOS. A tachometer disturbance were greatly suppressed. This demonstrates that
generator measured the rotational speed and converted it into our new method is very useful in rejecting disturbances and
a voltage. Next, a 12-bit A/D converter converted the analog improving control performance.
speed signal into a digital signal and sent it to the controller.
Then, the control input was calculated based on the output and
IV. C ONCLUSION
the tracking error between the reference speed and the output.
It and the disturbance were sent through a 12-bit D/A converter In a conventional servo system, the tracking performance
to the motor drivers. may be degraded by unknown disturbances even if an internal
388 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 55, NO. 1, JANUARY 2008

Fig. 11. Experimental system.

model of the reference input is employed. This paper describes


a new method of improving the disturbance-rejection perfor-
mance of a servo system that involves the estimation of an
EID and an improved servo system configuration based on
the method. This method has significant advantages over the
existing ones.
1) Even though the control system does not contain an inter-
nal model of a disturbance, the influence of an unknown
disturbance is almost completely rejected in both the
transient and steady-state responses.
2) The configuration is very simple.
3) It does not require the differentiation of measured
outputs.
4) It avoids cancellation of unstable poles/zeros.
5) The stability of the system can be broken down into two
independent parts: state feedback, and observer and low-
pass filter.
Guidelines for the selection of the low-pass filter are given,
based on an examination of simulation results, and a design
method for the state observer is presented, based on the concept
of perfect regulation. The validity of the method was demon-
strated using the speed control of a rotational control system as
an example.
It is also worth mentioning that the discussion in this paper
was restricted to the SISO case to keep the description simple.
However, the results obtained are easy to extend to the multi-
input multi-output case.

R EFERENCES
[1] M. Corless and J. Tu, “State and input estimation for a class of uncertain
systems,” Automatica, vol. 34, no. 6, pp. 757–764, 1998.
[2] M. Hou and R. J. Patton, “Optimal filtering for systems with unknown
inputs,” IEEE Trans. Autom. Control, vol. 43, no. 3, pp. 445–449,
Mar. 1998.
[3] J.-L. Chang, “Applying discrete-time proportional integral observers for
state and disturbance estimations,” IEEE Trans. Autom. Control, vol. 51,
no. 5, pp. 814–818, May 2006.
Fig. 12. Experimental results. (a) Disturbance. (b) Control results with- [4] F.-J. Lin and P.-H. Shen, “Robust fuzzy neural network sliding-mode
out disturbance estimation. (c) Control results with disturbance estimation. control for two-axis motion control system,” IEEE Trans. Ind. Electron.,
(d) Power spectral density. vol. 53, no. 4, pp. 1209–1225, Jun. 2006.
SHE et al.: IMPROVING DISTURBANCE-REJECTION PERFORMANCE BASED ON AN EID APPROACH 389

[5] M. Sun, Y. Wang, and D. Wang, “Variable-structure repetitive control: Mingxing Fang received the B.S. and M.S. degrees
A discrete-time strategy,” IEEE Trans. Ind. Electron., vol. 52, no. 2, in science from the Anhui Normal University, Wuhu,
pp. 610–616, Apr. 2005. China, in 1995 and 2003, respectively. He is cur-
[6] G. Cheng and K. Peng, “Robust composite nonlinear feedback control rently working toward the Ph.D. degree in control
with application to a servo positioning system,” IEEE Trans. Ind. Elec- engineering in the School of Information Science and
tron., vol. 54, no. 2, pp. 1132–1140, Apr. 2007. Engineering, Central South University, Changsha,
[7] K. Ohnishi, N. Matsui, and Y. Hori, “Estimation, identification, and sen- China.
sorless control in motion control system,” Proc. IEEE, vol. 82, no. 8, He has been working with the College of Physics
pp. 1253–1265, Aug. 1994. and Electronic Information, Anhui Normal Uni-
[8] S. Katsura, K. Ohnishi, and K. Ohishi, “Transmission of force sensation versity since 1995. His research interests include
by environment quarrier based on multilateral control,” IEEE Trans. Ind. mechatronics and the application of control theory.
Electron., vol. 54, no. 2, pp. 898–906, Apr. 2007.
[9] C.-S. Liu and H. Peng, “Inverse-dynamics based state and disturbance
observers for linear time-invariant systems,” Trans. ASME, J. Dyn. Syst.
Meas. Control, vol. 124, no. 3, pp. 375–381, 2002.
[10] Y. Xiong and M. Saif, “Sliding mode observer for nonlinear uncertain
systems,” IEEE Trans. Autom. Control, vol. 46, no. 12, pp. 2012–2017,
Dec. 2001.
[11] X. Chen, T. Fukuda, and K. D. Young, “A new nonlinear robust distur- Yasuhiro Ohyama (M’96) received the B.S., M.S.,
bance observer,” Syst. Control Lett., vol. 41, no. 3, pp. 189–199, 2000. and Ph.D. degrees in engineering from the Tokyo
[12] J.-H. She, Y. Ohyama, and M. Nakano, “A new approach to the estimation Institute of Technology, Tokyo, Japan, in 1980, 1982,
and rejection of disturbances in servo systems,” IEEE Trans. Control Syst. and 1985, respectively.
Technol., vol. 13, no. 3, pp. 378–385, May 2005. He worked on developing controllers for industrial
[13] W. S. Levine, The Control Handbook. Boca Raton, FL: CRC Press, robots and on CAD systems for control design as
1996. the Director of Advanced Control Laboratory, Inc.,
[14] L. R. Hunt, G. Meyer, and R. Su, “Noncausal inverses for linear systems,” Tokyo, from 1985 to 1991. He is currently a Profes-
IEEE Trans. Autom. Control, vol. 41, no. 4, pp. 608–611, Apr. 1996. sor with the School of Bionics, Tokyo University of
[15] T. Umeno, T. Kaneko, and Y. Hori, “Robust servosystem design with two Technology, Tokyo, where he does research on the
degrees of freedom and its application to novel motion control of robot application of control theory, robotics, and engineer-
manipulators,” IEEE Trans. Ind. Electron., vol. 40, no. 5, pp. 473–485, ing education.
Oct. 1993. Dr. Ohyama is a member of the Society of Instrument and Control Engineers
[16] S. Komada, N. Machii, and T. Hori, “Control of redundant manipulators and the Institute of Electrical Engineers of Japan.
considering order of disturbance observer,” IEEE Trans. Ind. Electron.,
vol. 47, no. 2, pp. 413–420, Apr. 2000.
[17] K. Zhou, J. C. Doyle, and K. Glover, Robust and Optimal Control.
Englewood Cliffs, NJ: Prentice-Hall, 1996.
[18] B. D. O. Anderson and J. B. Moore, Optimal Control—Linear Quadratic
Methods. Englewood Cliffs, NJ: Prentice-Hall, 1989.
[19] H. Kwakernaak and R. Sivan, “The maximally achievable accuracy of
linear optimal regulators and linear optimal filters,” IEEE Trans. Autom.
Control, vol. AC-17, no. 1, pp. 79–86, Feb. 1972. Hiroshi Hashimoto (M’93) received the Ph.D. de-
[20] H. Kimura, “A new approach to the perfect regulation and the bounded gree in engineering from Waseda University, Tokyo,
peaking in linear multivariable control systems,” IEEE Trans. Autom. Japan, in 1990.
Control, vol. AC-26, no. 1, pp. 253–270, Feb. 1981. He is currently an Associate Professor with the
[21] K. Sugiura and Y. Hori, “Vibration suppression in 2- and 3-mass system School of Bionics, Tokyo University of Technol-
based on the feedback of imperfect derivative of the estimated torsional ogy, Tokyo, where he does research on autonomous-
torque,” IEEE Trans. Ind. Electron., vol. 43, no. 1, pp. 56–64, Feb. 1996. distributed-cooperative robots, cybernetic interfaces,
[22] F. L. Lewis, Applied Optimal Control and Estimation—Digital Design and vision systems, welfare technology, and e-learning.
Implementation. Englewood Cliffs, NJ: Prentice-Hall, 1992. Dr. Hashimoto is a member of the Society of
[23] Signal Processing Toolbox User’s Guide Version 6, The MathWorks Inc., Instrument and Control Engineers and the Institute
Natick, MA, 2005. of Electrical Engineers of Japan.

Min Wu received the B.S. and M.S. degrees in


Jin-Hua She (M’94) received the B.S. degree engineering from the Central South University,
in engineering from the Central South University, Changsha, China, in 1983 and 1986, respectively. He
Changsha, China, in 1983, and the M.S. and Ph.D. received the Ph.D. degree in engineering from Tokyo
degrees in engineering from the Tokyo Institute Institute of Technology, Tokyo, Japan, in 1999.
of Technology, Tokyo, Japan, in 1990 and 1993, Since July 1986, he has been with the Central
respectively. South University, where he is currently a Professor
In 1993, he was with the Department of Mecha- of automatic control engineering in the School of
tronics, School of Engineering, Tokyo University Information Science and Engineering. He was a Vis-
of Technology, Hachioji, Tokyo, Japan, and in iting Scholar at the Department of Electrical En-
April 2004, he transferred to the University’s School gineering, Tohoku University, Sendai, Japan, from
of Bionics, where he is currently an Associate Pro- 1989 to 1990, and a Visiting Research Scholar at the Department of Control
fessor. His current research interests include the application of control theory, and Systems Engineering, Tokyo Institute of Technology, Tokyo, from 1996
repetitive control, process control, Internet-based engineering education, and to 1999. His current research interests are robust control and its application,
robotics. process control, and intelligent control.
Dr. She is a member of the Society of Instrument and Control Engineers Dr. Wu is a member of the Nonferrous Metals Society of China and the
and the Institute of Electrical Engineers of Japan. He received the Control China Automation Association. He received the Best Paper Award at the Inter-
Engineering Practice Paper Prize of the International Federation of Automatic national Federation of Automatic Control in 1999 (jointly with M. Nakano and
Control in 1999 (jointly with M. Wu and M. Nakano). J.-H. She).

You might also like