Loughborough University

Institutional Repository

Non-linear disturbance
observer-based robust control
for systems with mismatched
disturbances/uncertainties
This item was submitted to Loughborough University's Institutional Repository
by the/an author.
YANG, J., CHEN, W-H and LI, S., 2011. Non-linear disturbance
observer-based robust control for systems with mismatched disturbances/uncertainties.
IET Control Theory and Applications, 5 (18), pp. 2053 - 2062.
Citation:

Additional Information:

This article was published in the journal, IET Control Thec

ory and Applications [
Institution of Engineering and

Technology (IET)].The denitive version is available

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?arnumber=6086648

Metadata Record:
Version:

https://dspace.lboro.ac.uk/2134/11123

Accepted for publication

Publisher:

c Institution of Engineering and Technology

Please cite the published version.

at:

This item was submitted to Loughboroughs Institutional Repository

(https://dspace.lboro.ac.uk/) by the author and is made available under the
following Creative Commons Licence conditions.

For the full text of this licence, please go to:

http://creativecommons.org/licenses/by-nc-nd/2.5/

Nonlinear disturbance observer based robust control

for systems with mismatched
disturbances/uncertainties
Jun Yang, Wen-Hua Chen, and Shihua Li

Abstract:

Robust control of nonlinear systems with disturbances and uncertainties is addressed in

this paper using disturbance observer based control technique. In this framework, the disturbance
is a generalized concept, which may include external disturbances, unmodeled dynamics, and system parameter perturbations. The existing disturbance observer based control (DOBC) methods
were only applicable for the case where disturbances and uncertainties satisfy so called matching
condition, i.e., they enter the system in the same channel as the control inputs. By appropriately
designing a disturbance compensation gain vector in the composite control law, a nonlinear disturbance observer based robust control (NDOBRC) method is proposed in this paper to attenuate the
mismatched disturbances and the influence of parameter variations from system output channels.
The proposed method is applied to a nonlinear missile system in the presence of various uncertainties and external disturbances. Simulation shows that, compared with the widely used nonlinear
dynamic inversion control (NDIC) and NDIC plus integral action (NDIC+I) methods, the proposed
method provides much better disturbance attenuation ability and stronger robustness against various parameter variations. The proposed method significantly extends the applicability of the DOBC
methods.
Keywords:

Mismatched disturbances/uncertainties, nonlinear disturbance observer, disturbance

compensation gain, nonlinear missile.

1 Introduction
Disturbances including external disturbances, unmodeled dynamics, and parameter perturbations,
widely exist in aerospace engineering, such as aircrafts, missiles, and satellites, and also many other
engineering systems. Generally speaking, the control performance of these systems is severely affected by disturbances and uncertainties. Disturbance attenuation is of great importance in control
system design.
Disturbance observer based control (DOBC) provides a promising approach to handle system

J. Yang and S.H. Li are with School of Automation, Southeast University, Nanjing 210096, P.R. China,

emails: [email protected] and [email protected]

W.-H Chen is with Department of Aeronautical and Automotive Engineering, Loughborough University,
LE11 3TU, UK, email: [email protected]

disturbances and improve robustness [1, 2]. In this framework, a base line controller is firstly designed under the assumptions that there are no disturbances and uncertainties, and then a compensation is added to counteract the influence of disturbances and uncertainties that is estimated by a
properly designed disturbance observer (DOB) [3].
Compared with other robust control schemes, DOBC approach has two distinct features. One
feature is that disturbance observer based compensation can be considered as a patch for existing controllers that may provide good stability and tracking performance but have unsatisfactory
disturbance attenuation and robustness against uncertainties. The benefits of this is that there is no
change to the base line controller which may have been widely used and developed for many years
such as classical flight control systems. After the base line controller is designed using the existing
procedures, the disturbance observer based compensation is added to improve the robustness and
disturbance attenuation. Instead of employing a complete new and different control strategy which
demands a new verification and certification process, the verification of DOBC can be developed
based on the existing verification process to ensure safety and reliability. The second feature is that
DOBC is not a worst case based design. Most of the existing robust control methods are worst
case based design, and have been criticized as being over conservative. Promising robustness is
achieved with the price of degraded nominal performance. In DOBC approach, the nominal performance of the base line controller is recovered in the absence of disturbances or uncertainties.
Although in the worst cases for described uncertainties, DOBCs performance may be poorer than
that of other robust controllers, the overall better performance may be achieved as a dynamic system
more likely operates around its nominal condition.
Disturbance observer technique was firstly presented by Ohishi et al. [4] for a motion servo
system in the late of 1980s. During the past two decades, DOBC schemes for linear and nonlinear
systems have been successfully developed and applied in various engineering systems, such as servo
control system [2], robotic system [5], hard disk drive system [6], position system [7], grinding
system [8, 9], etc.
Note that the above mentioned DOBC schemes are generally designed according to linear control theory, even if the actual controlled plant is nonlinear [10]. In the presence of strong nonlinearities, the validity of using linear analysis and synthesis technique may be doubtful [11, 12, 13].
To this end, research of DOBC using nonlinear system theory has drawn much attention in the past
decade. Recent developments of nonlinear disturbance observer based control (NDOBC) can be
seen in [3, 11, 12, 13, 14, 15]. In [14], a sliding mode based nonlinear disturbance observer (NDOB)
was proposed for motor control. Using Lyapunov stability theory, Chen et al. [11] developed a
nonlinear disturbance observer for a single-input-single-output (SISO) nonlinear system subject to
unknown constant disturbance and applied it to a two-link manipulator. Rigorous exponential stability analysis of NDOBC was established based on Lyapunov stability theory in [3]. Guo and Chen
[12] proposed a DOBC method for a class of multi-input-multi-output (MIMO) nonlinear systems
2

with well-defined nonlinearity. Composite methods which combine DOBC with terminal sliding
model control or H control were proposed to solve the problem caused by disturbance model
perturbations in Ref. [12, 13, 15].
In the previous literature, the DOBC methods were only applicable to disturbances satisfying
matching condition [16]. Here matching condition means that the disturbances appear in the same
channels as the control inputs. For mismatched disturbances, i.e., the disturbances or the influence
of uncertainties enter the system in different channels from the control inputs, the existing DOBC
methods are not applicable. It should be pointed out that mismatched disturbances are more practical
and widely exist in engineering systems. Taking aircraft as an example, the lumped disturbance
torques caused by unmodeled dynamics, external winds, and parameter perturbations may influence
the states directly, rather than through the input channels.
The problem of mismatched disturbances have been concerned in [13, 15] where the matched
disturbances were canceled by DOBC while the mismatched disturbances were attenuated by H
control [15] or variable structure control [13]. It was also reported that some constrains with the
mismatched disturbances (such as with bounded H2 norm) were required in [13, 15]. In the presence
of mismatched disturbances/uncertainties, a widely used practical method is to add an integral action
in the feedback control law to remove offset of the closed-loop system [17], which will be compared
with the proposed method in this paper.
In this paper, a nonlinear disturbance observer based robust control (NDOBRC) method is proposed to solve the disturbance attenuation problem of nonlinear systems subject to mismatched
disturbances/uncertainties. In the presence of mismatched disturbances and uncertainties, it is unlikely to achieve asymptotic stability as the disturbance and the control inputs do not appear in the
same channels and the influence of the disturbances can not be completely canceled. Instead, the
design objective is to remove the influence of disturbances and uncertainties from the output. The
key issue here is how to design a disturbance compensation gain vector to assure that the mismatched
disturbances can be completely eliminated from the output channels. With the proposed NDOBRC
method, the prominent patch feature retains, the robustness and disturbance attenuation against a
much wider range of uncertainties and disturbances are significantly improved without sacrificing
the nominal performance of the original nonlinear control strategy.
To demonstrate the feasibility and effectiveness of the proposed method, control design of a
missile system using NDOBRC is investigated in this paper. The missile system considered is subject
to mismatched disturbances that include not only external disturbances, but also model uncertainties.
The rest of the paper are organized as follows. In Section 2, a preliminary regarding nonlinear
disturbance observers is presented for the convenience of the readers. The problem formulation is
then described. Design and analysis of the proposed NDOBRC is presented in Section 3. In Section
4, the proposed NDOBRC is applied to a nonlinear missile system. Simulation studies and results
are presented in Section 5. The conclusions are finally given in Section 6.
3

2 Preliminaries and problem formulation

2.1 Nonlinear disturbance observer
A general single-input-single-output (SISO) affine nonlinear systems with lumped disturbances is
represented as

x = f (x) + g1 (x)u + g2 (x)d,

y = h(x),

(1)

where x Rn , d Rn , u R and y R are the state vector, lumped disturbance vector, input
and output variables. It is assumed that f (x), g1 (x), g2 (x) and h(x) are smooth functions in terms
of state x.
For system (1), the following nonlinear disturbance observer (NDOB) is proposed in [1, 3, 18]
to estimate the unknown disturbances d, given by

d = z + p(x),
z = l(x)g (x)z l(x)[g (x)p(x) + f (x) + g (x)u],
2
2
1

(2)

where d and z are the estimates of the unknown disturbances and the internal states of the nonlinear
observer, respectively, and p(x) is a nonlinear vector-valued function to be designed. The nonlinear
observer gain l(x) is defined as
l(x) =

p(x)
.
x

(3)

Assumption 1: The lumped disturbance d is slowly time-varying, i.e., d 0.

It can be proved that, under the assumption that the disturbances are slowly time-varying, d
approaches d asymptotically if p(x) is chosen such that
+
e(t)

p(x)
g2 (x)e(t) = 0,
x

(4)

is globally stable for all x Rn , where the estimation error is defined as

e = d d.

(5)

Clearly, any nonlinear vector-valued function l(x) which makes Eq. (4) asymptotically stable
can be chosen to guarantee the asymptotic convergence of the estimation error.
Remark 1: The rigorous asymptotic convergence of NDOB has been established under the condition that the disturbances vary slowly relative to the observer dynamics (i.e., Assumption 1). It has
been also reported that the observer (2) can track some fast time-varying disturbances with bounded
error as long as the derivative of the disturbances is bounded [11].
Remark 2: In the presence of uncertainties, the lumped disturbances would be a function of the
states, which can be reasonably estimated if the disturbance observer dynamics is faster than the
closed-loop dynamics. The same argument for the state observer based control methods is applicable.
4

2.2 Problem formulation

In the previous literature, the DOBC methods only deal with the case of matched disturbances, i.e.,
the lumped disturbances d enter the system with the same channels as the control inputs. Precisely
speaking, the matched disturbances means that the following two conditions are satisfied: (i) the
control inputs u and the lumped disturbances d have the same dimension, and (ii) g1 (x) = g2 (x)
in Eq. (1). These conditions have restricted the application of DOBC strategies to more general
controlled plants.
Remark 3: Note that the NDOB (2) is applicable for the case of mismatched disturbances.
However, the estimates of NDOB can not be used to compensate the disturbances directly because
the disturbances are not in the same channels with the control inputs.
It should be pointed out that in general, the influence of the mismatched disturbances can not
be removed from state variables. In this paper, based on the disturbance estimate of NDOB (2), the
composite control law as u = (x) + (x)d is designed to remove the influence of the lumped
disturbance from the output channel by appropriately designing the compensation gain (x). This
will substantially extends the application fields of the DOBC strategy.
A general design procedure of nonlinear disturbance observer based robust control (NDOBRC)
for system (1) subject to mismatched disturbances is proposed as follows:
(i) Design a base line nonlinear feedback controller to achieve stability and performance specifications without taking into account disturbances/uncertainties.
(ii) Lump the external disturbances and the influences of the uncertainties, and then design a nonlinear disturbance observer to estimate the lumped disturbances.
(iii) Design a disturbance compensation gain vector to achieve desired performance specification in
the presence of external disturbances and uncertainties.
(iv) Construct the composite NDOBRC law by integrating the nonlinear feedback controller and
the disturbance observer based compensation part.

3 NDOBRC for nonlinear systems with mismatched disturbances

3.1 Composite control law
For the nonlinear systems (1) with mismatched lumped disturbances, the composite control law of
NDOBRC is designed as

u = (x) + (x)d,

(6)

where (x) is the feedback control law without considering the disturbances, (x) is the disturbance
compensation gain vector to be designed, and d is the disturbance estimate based on the NDOB (2).
Obviously, the composite control law (6) consists of two parts: one is the nonlinear feedback
control part (x), and the other is the disturbance compensation term (x)d based on the NDOB.
5

Note that in the composite control law (6), the disturbance compensation term (x)d is just
designed for disturbances, i.e., the NDOB works if and only if disturbances exist. Thus it just
works like a patch for the existing controller to improve its disturbance attenuation and robustness
against uncertainties. In the absence of disturbances and uncertainties, the nominal performance of
the existing controller recovers.

3.2 Stability analysis

To establish the stability of the closed-loop system, the following preliminary result is required.
Lemma 1: Consider a nonlinear system x = H(x, d) with state x Rn , input d Rm , in
which H(0, 0) = 0. Suppose the equilibrium x = 0 of x = H(x, 0) is globally asymptotically
stable. Then there exists an m m matrix M (x) of smooth functions of x, which is defined for
all x Rn and is nonsingular for all x, such that x = H(x, M (x)d) is input-to-state stable (ISS)
[19].
Consider the lumped disturbances d as the inputs of the closed-loop system, also the system
state x and observer state e as of the state of the closed-loop system. Let

=
x

=
F (x)

x
e

(7)

f (x) + g1 (x)(x) g1 (x)(x)e

l(x)g2 (x)e

(8)

The input-to-state stability of the closed-loop system is established by the following theorem.
Theorem 1: The closed-loop system consists of nonlinear system (1), composite control law (6)
and nonlinear disturbance observer (2) is ISS if the following conditions are satisfied:
(i). the nonlinear system (1) under the original designed controller u = (x) is globally asymptotically stable in the absence of disturbances,
(ii). the vector-valued function p(x) is chosen such that the observer error system (4) is globally
asymptotically stable,
(iii). there exist a disturbance compensation gain (x) such that

g2 (x) + g1 (x)(x) = G(x)M

(x),

(9)

is an arbitrary matrix and M (x)

is a matrix such that
holds, where G(x)

= F (x)
+
x
is ISS.
6

G(x)M
(x)
0

d,

(10)

Proof: Combining system (1), composite control law (6), disturbance estimation error function
(4) and (5) together, the closed-loop system is obtained

x = [f (x) + g1 (x)(x)] g1 (x)(x)e + [g2 (x) + g1 (x)(x)]d,

p(x)

e =
g2 (x)e.
x
Combining Eqs. (7), (8) with (11), the closed-loop system is given as

g2 (x) + g1 (x)(x)
d.
= F (x)
+
x
0

(11)

(12)

= F (x)
is asymptotically stable
With the conditions given in (i) and (ii), it can be shown that x
in the absence of disturbances. This means that there exists a C 1 positive definite and proper function
and a class K function () such that
V (x)
x Rn /0

V
< (kxk),
F (x)

(13)

where kk denotes Euclidean norm of a vector, definitions of class K and K functions can be found
in [19](pp. 1).
Considering the following system

= H (x,
d) = F (x)
+
x

G(x)
0

d.

(14)

Combining (13) with (14) gives

x Rn /0

V
0) < (kxk).
H(x,

(15)

Similar with the proof of Lemma 1 in [19](pp. 27-28), it can be shown that there exists a n n
matrix M (x) which is defined for all x Rn and is nonsingular for all x, such that for some class
K function ()
kxk (kdk)

V
1
M (x)d)

H (x,
< (kxk).

x
2

(16)

Consider the condition given in (iii) and (16), it can be concluded that the closed-loop system
(12) is ISS. 
3.3 Design of the disturbance compensation gain
The nonlinear systems (1) and the composite control law (6) can be rewritten and expressed as

x = f(x)x + g (x)u + g2 (x)d,

1
(17)
y = h(x)x,

and

u = (x)x

+ (x)d,

where f(x) = f (x)/x, h(x)

= h(x)/x, and (x)

= (x)/x.
7

(18)

Theorem 2: Suppose Assumption 1 is satisfied. Consider nonlinear system (1) under the composite control law (6) consisting of the nonlinear feedback control law (x) and the disturbance
compensation term (x)d based on the estimates of the NDOB (2). The influence of the lumped
disturbances can be eliminated from the output channel in steady-state if the nonlinear disturbance
compensation gain (x) is selected such that
(i). the closed-loop system (11) is ISS.
(ii). the following condition holds
n
o1

1

(x) = h(x)
f(x) + g1 (x)(x)

g1 (x)

1

g2 (x).
h(x)
f(x) + g1 (x)(x)

(19)

Proof: Considering the closed-loop system (11) with Eqs. (17) and (18), the states can be
expressed as

1
{x g1 (x)(x)e [g1 (x)(x) + g2 (x)] d}.
x = f(x) + g1 (x)(x)

(20)

Combining Eqs. (19), (20) with the output equation in (17) yields

1

1

y = h(x)
x + h(x)

g2 (x)e.
f(x) + g1 (x)(x)
f(x) + g1 (x)(x)

(21)

0
Since the closed-loop system is stable, the following two conditions are satisfied, i.e., lim x(t)
t

and lim e(t) 0. The later follows from the properly designed disturbance observers. It can be
t

shown that the disturbances can be finally attenuated from the output in steady-state as y = 0. 
Remark 4: Let g2 (x) = (g21 (x), g22 (x), , g2n (x)), a more explicit expression of the disturbance compensation gain can be given as
(x) = (1 (x), 2 (x), , n (x)) ,

(22)

where

i (x) =

f(x) + g1 (x)(x)

h(x)

f(x) + g1 (x)(x)

g1 (x)

h(x)

det

det

g2i (x)

In fact,

1

h(x)
f(x) + g1 (x)(x)

g1 (x)

=
similarly, for all i = 1, 2, , n,

h(x)adj
f(x) + g1 (x)(x)

g1 (x)


det f(x) + g1 (x)(x)

f(x) + g1 (x)(x)

g1 (x)

det

h(x)
0


,
det f(x) + g1 (x)(x)


1

h(x)
f(x) + g1 (x)(x)

g2i (x) =

det

f(x) + g1 (x)(x)

g2i (x)

h(x)
0


det f(x) + g1 (x)(x)

(23)

(24)
,

Combining Eqs. (19), (23) and (24), yields the result of Remark 4.
Remark 5: Note that the nonlinear disturbance compensation gain vector (x) in (19) is a
general case and suitable for both matched and mismatched disturbances. In the matched case, i.e.,
g1 (x) = g2 (x), it can be obtained from (19) that the nonlinear disturbance compensation gain
vector reduces to (x) = 1 which is widely used in the previous DOBC designs [1, 12, 13, 15].

4 Control design of a nonlinear missile

4.1 Longitudinal dynamics of a missile system
The model of the longitudinal dynamics of a missile under consideration is taken from Refs. [1, 20],
described by
= f1 () + q + b1 () + d1 ,

(25)

q = f2 () + b2 + d2 ,

(26)

where is the angle of attack (degrees), q is the pitch rate (degrees per second), and is the tail fin
deflection (degrees). The disturbances d1 and d2 denote the lumped disturbance torques which may
be caused by unmodeled dynamics, external wind, and variation of aerodynamic coefficients, etc.
The nonlinear functions f1 (), f2 (), b1 (), and b2 are determined by aerodynamic coefficients.
When the missile travels at Mach 3 at an altitude of 6,095 m (20,000 ft) and the angle of attack
|| 20 deg, they are given by
f1 () =

180gQS
W V

f2 () =

cos( 180
)(1.03 104 3 9.45 103 || 1.7 101 ),

180QSd
Iyy (2.15

104 3 1.95 102 || + 5.1 102 ),

180gQS
cos(
),
W V
180
180QSd
b2 = 0.206
.
Iyy

b1 () = 3.4 102

(27)
(28)
(29)
(30)

The tail fin actuator dynamics are approximated by a first-order lag process, i.e.,
= (1/t1 )( + u) + d3 ,

(31)

where u the commanded fin defection (degrees), d3 the disturbance which may influence the actuator
dynamics (e.g. frictions) and t1 the time constant (seconds). The physical meaning and values of the
parameters in Eqs. (27)-(31) for the missile under consideration are listed in Table 1.
4.2 Nonlinear dynamic inversion control
In the absence of disturbances d1 , d2 and d3 , an autopilot for the missile to track an angle-of-attack
reference (t) may be designed using the nonlinear dynamic inversion control (NDIC) [21]. The
output is chosen as
y = + kq q,
9

(32)

Table 1: Parameters in longitudinal dynamics of the missile

Parameter

Symbol

Value

Weight

4,410 kg

Velocity

947.6 m/s

Pitch moment of inertia

Iyy

247.44 kg m2

Dynamic pressure

293,638 N/m2

Reference area

0.04087 m2

Reference diameter

0.229 m

Gravitational acceleration

9.8 m/s2

Time constant of tin actuator

t1

0.1 s

where kq is a chosen constant. The resultant control law is given by

undic = [t1 /(b1 + kq b2 )]{k1 (y )

(33)

+ k2 [f1 + q + b1 + kq (f2 + b2 ) ]
+m
},
where
m = m1 [f1 () + q + b1 ()] + f2 () + b2 ,
m1 =

f2 ()
f1 () b1 ()
+
+ kq
,

(34)
(35)

and k1 , k2 are constant gains to be designed according to desired closed-loop behaviors.

Suppose that the command signal cmd is filtered by a low-pass prefilter to provide the reference
for tracking
G(s) = n2 /(s2 + 2n s + n2 ).

(36)

Substituting the NDIC law (33) into the longitudinal dynamics of the missile, the closed-loop
error dynamics are given by
y(t)
(t) + k2 [y(t)
(t)]

+ k1 [y(t) (t)] = 0.

(37)

In this paper, the parameters in Eqs. (32)-(37) are chosen as

= 0.7, n = 10(rad/s),

(38)

kq = 0.06(s),

(39)

k1 = 15(1/s2 ), k2 = 6(1/s).

(40)

As shown in Eq. (37), the longitudinal dynamics of the missile is feedback linearized by NDIC.
The closed-loop poles under NDIC is given by 7.0 7.14j, thus promising tracking performance
is achieved under the control law in the absence of disturbances. However, it is reported that such
NDIC scheme has poor robustness and disturbance rejection ability [1, 22].
10

4.3 NDOBRC for the missile system

Define the state vector of the missile system as x = [, q, ]T . Rearranging the state equations of the
missile systems (25), (26), (31) and the output equation (32), the formulation like affine nonlinear
systems (1) is obtained. The concrete nonlinear vector-valued functions are denoted as

f1 () + q + b1 ()

f (x) =
,
f2 () + b2

(1/t1 )

g1 (x) =

0
1/t1

1 0

g2 (x) = 0 1

0 0

(41)

(42)

0 ,

(43)

h(x) = + kq q.

(44)

The lumped disturbance vector is

d1

d = d2 .

d3

(45)

Using the nonlinear disturbance observer (2) introduced in Section 2, the lumped disturbances
in the missile system can be estimated.
To determine the nonlinear disturbance compensation gain vector for the missile system following from the procedure proposed in Section 3, the above dynamics are reformulated as Eq. (17). The
nonlinear matrix-value functions g1 (x) and g2 (x) are the same as those in Eqs. (42) and (43), while

f(x) and h(x)

are denoted as follows

f1 ()/

f(x) = f2 ()/

h(x)
=

1,

b1 ()

b2

0 1/t1
kq ,

(46)

(47)

Rearranging the NDIC law (33), gives

undic = (x)x

+ (, ,

),
where (x)

= [
1 (x),
2 (x),
3 (x)], and


f1 ()
f2 ()
t1
,
k1 + (k2 + m1 )
+(1 + k2 kq )

1 (x) =
b1 + k q b2

11

(48)

(49)


2 (x) =

3 (x) = 1

t1
(k1 kq + k2 + m1 ),
b1 + k q b2

(50)

t1
(k2 b1 + k2 kq b2 + m1 b1 + b2 ),
b1 + k q b2

(51)

t1
(k1 + k2 +
).
b1 + k q b2

(52)

(, ,

) =

Since all nonlinear functions including f(x), h(x),

g1 (x), g2 (x) and (x)

have been obtained,

the nonlinear disturbance compensation gain vector (x) can be obtained by using Eq. (19).
The composite nonlinear disturbance observer based robust control (NDOBRC) law is given by

u = undic + (x)d,

(53)

where undic is the NDIC law (33), (x) is the disturbance compensation vector given by (19), and d
is disturbance estimate governed by (2). The control structure of the nonlinear disturbance observer
based NDIC scheme is shown in Fig. 1.
d

cmd

Prefilter

NDIC

u ndic

External
Disturbances

Nonlinear System
with Uncertainties

! (x)
d
p (x)
+

"

Nonlinear
Equation
NDOB

Figure 1: Control structure of nonlinear disturbance observer based robust control (NDOBRC)
scheme.

5 Simulation studies
In this paper, the nonlinear vector-valued function l(x) in NDOB (2) is chosen as

10

l(x) = 0

0
10
0

0 .

10

(54)

It is reported that the NDIC method has a poor disturbance rejection ability and robustness [1].
Integral action is a practical and most widely used method to eliminate the steady-state error in
the presence of disturbances/uncertainties [17]. To demonstrate the effectiveness of the proposed
method, in addition to the NDIC method, NDIC plus an integral action (called NDIC+I) is also
12

employed for comparison. The control law of the NDIC+I is represented as follows
R
undic+i = [t1 /(b1 + kq b2 )]{k0 (y )d + k1 (y )

(55)

+ k2 [f1 + q + b1 + kq (f2 + b2 ) ]
+m
},

where the integral coefficient is chosen as k0 = 20 to achieve a satisfactory performance.

It can be seen from the control law (55) that the NDIC+I method has a PID like structure.
Furthermore, the closed-loop system under Eq. (55) is given by
y
+ k2 (y )
+ k1 (y ) + k0

(y )d = 0.

(56)

As will be shown by simulation later, the offset caused by disturbances and uncertainties can be
eliminated by the NDIC+I method.
5.1 External disturbance rejection ability
In this subsection, the external disturbance rejection ability of the missile system under the proposed
NDOBRC method is investigated. Considering the case that the external disturbances d1 = 1 is
imposed on systems at t = 6 sec while d2 = 5 enters at t = 2 sec, the response curves of both
the output and input under three control methods are shown in Fig. 2. The corresponding response
curves of the states are shown in Fig. 3.
It can be observed from Fig. 2(a) that the NDIC method results in a large steady-state error. As
for the NDIC+I method, there is no steady-state error any more, but large overshoot and long settling
time are experienced. The proposed NDOBRC exhibits a much better transient and steady-state
performance, such as small overshoot, short settling time and zero steady-state error. The control
profile in Fig. 2(b) shows that no excessive (or) high gain is demanded for all the three methods.
Fig. 3 shows that all states under the three methods remain within allowable regions. This
means that the proposed method gains much better external disturbance rejection performance than
the other two methods without bringing adverse effects to all the states.
5.2 Robustness against model uncertainties
The robustness against model uncertainties of the proposed NDOBRC is tested in this part. To
investigate the performance of robustness in detail, three cases of model uncertainties are considered.
Case I: both f1 () and f2 () have variations of +20%.
For the first case of model uncertainties, the response curves of the output/input and the states
under the three control methods are shown in Figs. 4 and 5, respectively. It can be observed from
Fig. 4(a) that the NDIC scheme has resulted in large steady-state error. For the NDIC+I method, the
steady-state error was eliminated but quite slowly. The proposed NDOBRC method approaches to
the reference setpoint rapidly without steady-state error.
As shown in Figs. 4(b) and 5, both the control input and the states remain within allowable
13

12
Reference
NDOBRC
NDIC
NDIC+I

10

Output, y (deg)

8
6
4
2
0
2
0

(a)

4
Time,sec

Commanded fin deflections, u (deg)

10
NDOBRC
NDIC
NDIC+I

10

15
0

(b)

4
Time,sec

Figure 2: Response curves of the output and input in the presence of external disturbances under the
control laws of NDOBRC (53) (solid line), NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted
line). The reference signal is denoted by dotted line.

regions. These variables under the NDOBRC method converge to the desired equilibrium point
much quicker than those under the other two methods.
Case II: f1 () and f2 () have variations of -20% and -5%, respectively.
In such case, Figs. 6 and 7 show the response curves of the output/input and states under the
three control methods. It can be observed from Fig. 6(a) that the output under the NDIC method
substantially departs from the desired reference trajectory. Essentially, this implies that the missile
becomes unstable as the model is only valid when || 20 degrees, but the angle of attack reaches
70 degrees in simulation. Fig. 6(a) shows that the output under the NDIC+I method asymptotically approaches to the setpoint with oscillation and large overshoot. The proposed NDOBRC has
achieved the best performance, including a small overshoot, a short settling time, no oscillation and
zero steady-state error.
As shown in Fig. 6(b), the magnitude of the control input under the NDOBRC is much smaller
14

than those under the other two methods. The NDIC method is unusable in this case because the
magnitude of the control input is huge and over the actuator constraint. Also from the response
curves of the states in Fig. 7, it can be concluded that the proposed method gains the best performance
of robustness among all the three methods.
Case III: f1 ()and f2 () have variations of -20% and -7%, respectively.
The case of an even more severe model uncertainties is investigated in this part. The response
curves of the output/input and states are shown in Figs. 8 and 9, respectively.
In this case, the control performances under the NDOBRC and NDIC are quite similar with
those in Case II. However, the response curves under the NDIC+I method becomes oscillating and
unstable. This shows that the proposed NDOBRC achieves much better robust performance and
stability compared with other two methods.

6 Conclusion
As clearly demonstrated in the missile example in this paper, external disturbances, unmodeled dynamics and parameter perturbations always bring adverse effects to stability and performance of control systems. Disturbance observer based control (DOBC) has provided a solution to this problem.
It can significantly improve disturbance attenuation ability and robustness against uncertainties, and
acts like a patch to the existing design without considerably changing the nominal control design.
Existing DOBC methods were only applicable to matched disturbances. To this end, a nonlinear
disturbance observer based robust control (NDOBRC) method has been proposed for nonlinear systems in the presence of mismatched disturbances and uncertainties. It is shown that by appropriately
design the nonlinear compensation gains, zero steady-state tracking error can be achieved on system
output. Simulation studies of a missile system have been carried out to demonstrate the validity of
the proposed NDOBRC method. The results have shown that the proposed method obtains much
better disturbance rejection ability and robustness against model uncertainties as compared with
nonlinear dynamic inversion control (NDIC) and nonlinear dynamic inversion control plus integral
action (NDIC+I) methods.

Acknowledgements
This work was supported in part by Natural Science Foundation of Jiangsu Province (BK2008295),
Science Foundation of Aeronautics (20080769006), Graduate Innovation Program of Jiangsu
Province (CX10B 077Z), and Scientific Research Foundation of Graduate School of Southeast University. J. Yang would like to thank the China Scholarship Council (CSC) for supporting his visiting
in the UK. He also acknowledges the Department of Aeronautical and Automotive Engineering for
hosting his research at Loughborough University.
15

References
[1] Chen, W.-H.: Nonlinear disturbance observer-enhanced dynamic inversion control of missiles, J. Guid. Control Dyn., 2003, 26, (1), pp. 161-166
[2] Yao, B., Majed, M.A., and Tomizuka, M.: High-performance robust motion control of machine tools: an adaptive robust control approach and comparative experiments, IEEE/ASME
Trans. Mechatron., 1997, 2, (2), pp. 63-76
[3] Chen, W.-H.: Disturbance observer based control for nonlinear systems, IEEE/ASME Trans.
Mechatron., 2004, 9, (4), pp. 706-710
[4] Ohishi, K., Nakao, M., Ohnishi, K., and Miyachi, K.: Microprocessor-controlled DC motor
for load-insensive position servo system, IEEE Trans. Ind. Electron., 1987, 34, (1), pp. 44-49
[5] Komada, S., Machii, N., and Hori, T.: Control of redundant manipulators considering order of
disturbance observer, IEEE Trans. Ind. Electron., 2000, 47, (2), pp. 413-420
[6] Ishikawa, J., and Tomizuka, M.: A novel add-on compensator for cancellation of pivot nonlinearities in hard disk drives, IEEE Trans. Magn., 1998, 34, (4), pp. 1895-1897
[7] Kempf, C.J., and Kobayashi, S.: Disturbance observer and feedforward design for a highspeed direct-drive position table, IEEE Trans. Control Syst. Technol., 1999, 7, (5), pp. 513-526
[8] Chen, X.S., Yang, J., Li, S.H., and Li, Q.: Disturbance observer based multi-variable control
of ball mill grinding circuits, J. Process Control, 2009, 19, (7), pp. 1205-1213
[9] Yang, J., Li, S.H., Chen, X.S., and Li, Q.: Disturbance rejection of ball mill grinding circuits
using DOB and MPC, Powder Technol., 2010, 198, (2), pp. 219-228
[10] Yang, Z.-J., Tsubakihara, H., Kanae, S., Wada, K., and Su, C.-Y.: A novel robust nonlinear
motion controller with disturbance observer, IEEE Trans. Control Syst. Technol., 2008, 16,
(1), pp. 137-147
[11] Chen, W.-H., Ballance, D.J., Gawthrop, P.J., and OReilly, J.: A nonlinear disturbance observer for robotic manipulators, IEEE Trans. Ind. Electron., 2000, 47, (4), pp. 932-938
[12] Guo, L., and Chen, W.-H.: Disturbance attenuation and rejection for systems with nonlinearity
via DOBC approach, Int. J. Robust Nonlinear Control, 2005, 15, (3), pp. 109-125
[13] Wei, X.J., and Guo, L.: Composite disturbance-observer-based control and terminal sliding
mode control for non-linear systems with disturbances, Int. J. Control, 2009, 82, (6), pp.
1082-1098
[14] Chen, X.K., Komada, S., and Fukuda, T.: Design of a nonlinear disturbance observer, IEEE
Trans. Ind. Electron., 2000, 47, (2), pp. 429-436
[15] Wei, X.J., and Guo, L.: Composite disturbance-observer-based control and H-infinity control
for complex continuous models, Int. J. Robust Nonlinear Control, 2010, 20, (1), pp. 106-118
[16] Barmish, B., and Leitmann, G.: On ultimate boundedness control of uncertain systems in the
absence of matching assumptions, IEEE Trans. Autom. Control, 1982, 27, (1), pp. 153-158
[17] Lin, C.-F., Cloutier, J.R., and Evers, J.H.: High-performance, robust, bank-to-turn missile
autopilot design, J. Guid. Control Dyn., 1995, 18, (1), pp. 46-53
[18] Chen, W.-H., Ballance, D.J., Gribble, J.J., and OReilly, J.: Nonlinear PID predictive controller, IEE Proc.-Control Theory Appl., 1999, 146, (6), pp. 603-611
[19] Isidori, A.: Nonlinear Control Systems II (Springer-Verlag, London, 1999)
[20] Lu, P.: Nonlinear predictive controllers for continuous systems, J. Guid. Control Dyn., 1994,
17, (3), pp. 553-560
[21] Lane, S.H., and Stengel, R.F.: Flight control design using non-linear inverse dynamics, Automatica, 1988, 24, (4), pp. 471-483
[22] Schumacher, C., and Khargonekar, P.P.: Missile autopilot designs using H control with gain
scheduling and dynamic inversion, J. Guid. Control Dyn., 1998, 21, (2), pp. 234-243

16

12
NDOBRC
NDIC
NDIC+I

Angle of attack, (deg)

10
8
6
4
2
0
2
0

(a)

4
Time,sec

40
NDOBRC
NDIC
NDIC+I

Pitch rate, q (deg/sec)

30
20
10
0
10
20
30
40
0

(b)

4
Time,sec

Tail fin deflection, (deg)

0
1
2
3
4
5
NDOBRC
NDIC
NDIC+I

6
7
0

(c)

4
Time,sec

Figure 3: Response curves of the states in the presence of external disturbances under the control
laws of NDOBRC (53) (solid line), NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted line).

17

12

Output, y (deg)

10

4
Reference
NDOBRC
NDIC
NDIC+I

0
0

(a)

10
Time,sec

15

20

Commanded fin deflections, u (deg)

2
0
2
4
6
8
10
NDOBRC
NDIC
NDIC+I

12
14
0

(b)

10
Time,sec

15

20

Figure 4: Response curves of the output and input in the first case of model uncertainties (both f1 ()
and f2 () have variations of +20%) under the control laws of NDOBRC (53) (solid line), NDIC (33)
(dashed line), and NDIC+I (55) (dash-dotted line). The reference signal is denoted by dotted line.

18

10

Angle of attack, (deg)

NDOBRC
NDIC
NDIC+I

0
0

(a)

10
Time,sec

15

20

40
NDOBRC
NDIC
NDIC+I

Pitch rate, q (deg/sec)

30

20

10

10

20
0

(b)

10
Time,sec

15

20

0
NDOBRC
NDIC
NDIC+I

Tail fin deflection, (deg)

1
2
3
4
5
6
7
0

(c)

10
Time,sec

15

20

Figure 5: Response curves of the states in the first case of model uncertainties (both f1 () and f2 ()
have variations of +20%) under the control laws of NDOBRC (53) (solid line), NDIC (33) (dashed
line), and NDIC+I (55) (dash-dotted line).

19

70
60

Output, y (deg)

50
Reference
NDOBRC
NDIC
NDIC+I

40
30
20
10
0
0

10
Time,sec

(a)

15

20

Commanded fin deflections, u (deg)

20

20
NDOBRC
NDIC
NDIC+I

40

60

80

100
0

10
Time,sec

(b)

15

20

Figure 6: Response curves of the output and input in the second case of model uncertainties (f1 ()
and f2 () have variations of -20% and -5%, respectively) under the control laws of NDOBRC (53)
(solid line), NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted line). The reference signal is
denoted by dotted line.

20

70

Angle of attack, (deg)

60
50
40

NDOBRC
NDIC
NDIC+I

30
20
10
0
0

10
Time,sec

(a)

15

20

50
NDOBRC
NDIC
NDIC+I

Pitch rate, q (deg/sec)

40

30

20

10

10
0

(b)

10
Time,sec

15

20

20

Tail fin deflection, (deg)

20
NDOBRC
NDIC
NDIC+I

40

60

80

100
0

10
Time,sec

(c)

15

20

Figure 7: Response curves of the states in the second case of model uncertainties (f1 () and f2 ()
have variations of -20% and -5%, respectively) under the control laws of NDOBRC (53) (solid line),
NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted line).

21

100

Output, y (deg)

80
Reference

60

NDOBRC
NDIC

40

NDIC+I

20

20
0

10

10

Time,sec

(a)

Commanded fin deflections, u (deg)

20

20

40

NDOBRC
NDIC
NDIC+I

60

80

100
0

6
Time,sec

(b)

Figure 8: Response curves of the output and input in the third case of model uncertainties (f1 ()
and f2 () have variations of -20% and -7%, respectively) under the control laws of NDOBRC (53)
(solid line), NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted line). The reference signal is
denoted by dotted line.

22

100

Angle of attack, (deg)

80
NDOBRC

60

NDIC
NDIC+I

40

20

20
0

10

Time,sec

(a)
100

NDOBRC
NDIC
NDIC+I

Pitch rate, q (deg/sec)

80
60
40
20
0
20
40
60
0

10

Time,sec

(b)
20

Tail fin deflection, (deg)

20

40

60
NDOBRC
NDIC
NDIC+I

80

100
0

10

Time,sec

(c)

Figure 9: Response curves of the states in the third case of model uncertainties (f1 () and f2 ()
have variations of -20% and -7%, respectively) under the control laws of NDOBRC (53) (solid line),
NDIC (33) (dashed line), and NDIC+I (55) (dash-dotted line).

23