This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING 1

Disturbance Observer-Based Adaptive Tracking

Control With Actuator Saturation and Its Application
Huihui Pan, Weichao Sun, Member, IEEE, Huijun Gao, Fellow, IEEE, and Xingjian Jing, Member, IEEE

Abstract—This paper is concerned with the problem of adap- I. INTRODUCTION

tive tracking control for a class of nonlinear systems with para-
metric uncertainty, bounded external disturbance, and actuator
saturation. In order to achieve robust output tracking for the sat-
urated uncertain nonlinear systems, a combination of adaptive ro-
bust control (ARC) and a novel terminal sliding-mode-based non-
A DAPTIVE tracking control is popular in engineering and
science [1], [2]. However, it still encounters many im-
portant challenges, such as designers are likely to encounter a
linear disturbance observer (TSDO) is proposed, where the mod- lot of nonlinearities and modeling uncertainties, which refer to
eling inaccuracy and disturbance are integrated as a lumped dis-
turbance. Specifically, the observer errors of estimating the lump structured uncertainties (unknown parameters combined with
disturbances converge to zero in finite-time for improving the pre- known basis functions) and unstructured uncertainties (unmod-
cision of estimation. The estimated disturbances are then used in eled nonlinearities such as saturation, dead zone and backlash).
the controller to compensate for the system’s lumped disturbances. These nonlinearities and uncertainties could severely deterio-
The analytical results show that the proposed scheme is stable and rate the achievable control performance, leading to undesirable
can guarantee the asymptotic tracking with the tracking error con-
verging to zero even in the presence of disturbances. Finally, the de- control accuracy, limit cycles, and even instability.
veloped method is illustrated the effectiveness by the application to In order to deal with the nonlinear dynamics, a feedback lin-
control of a quarter-car model with active suspension system. earization (FL) approach has been proposed for designing the
trajectory tracking controllers for nonlinear systems [3], [4]. It
Note to Practitioners—Tracking problem is a popular topic in the has been shown that the nonlinear model is globally feedback
field of control, which plays an important role in actual application.
Especially this issue on nonlinear system, which dynamic behaviors linearizable through nonlinear coordinate transformation and
are complex and nonlinear characteristic property and the mathe- static state feedback, and therefore, globally stable controllers
matical models are often difficult to accurate obtain. Another chal- can be derived using the classical geometric techniques. While
lenging issue in nonlinear control is the effects from uncertainties the FL-based controller offers highly satisfactory performance
and external disturbances. Usually, building a dynamical system in ideal scenario, it may suffer from certain drawbacks when
called an “disturbance observer” to estimate the modeling inac-
curacy and disturbance is more applicable since uncertainties and put into practice. In general, the FL is a pure model-based con-
disturbance cannot be measured through sensors. In addition, it is trol, which is designed based on a nominal model of physical
noted that saturation nonlinearity occurs in many practical actu- systems and requires exact cancellation of nonlinearities. In re-
ators. How to design the controller in presence of saturation has ality, these conditions are hard to meet, because there always
become one main focus of control researches. Motivated by these exist parameter deviation and unmodeled nonlinearities which
issues, in this paper, ARC with TSDO algorithms are proposed for
a class of uncertainty nonlinear systems to avoid employing large cannot be modeled by explicit functions. Therefore, the FL con-
feedback gains in practical implementations. trol laws may not offer satisfactory performance.
In order to overcome FL controller’s disadvantage, a great
Index Terms—Adaptive robust control, active suspension sys- deal of effort has been devoted to address this issue, and con-
tems, disturbance observer, saturation nonlinearity. sequently, a number of methods/approaches has been proposed
to robustify systems in the presence of uncertainties and distur-
Manuscript received September 22, 2014; revised January 09, 2015; accepted bances. For example, to handle both structured and unstructured
February 12, 2015. This paper was recommended for publication by Associate uncertainties together in one controller, adaptive robust (ARC)
Editor Y. Zhang and Editor J. Wen upon evaluation of the reviewers' com-
proposed by Yao [5], [6]. These ARC controllers guaranteed
ments. This work was supported in part by the National Natural Science Foun-
dation of China under Grant 61403098 and Grant 61374041, in part by the Self- prescribed output tracking transient performance and final
Planned Task (No. SKLRS201308B) of the State Key Laboratory of Robotics tracking accuracy, in general, while achieving asymptotic
and System (HIT), and in part by Internal Research Funds of the Hong Kong
output tracking in the absence of unstructured uncertainties.
Polytechnic University.
H. Pan and W. Sun are with the State Key Laboratory of Robotics and Although ARC controllers can achieve excellent tracking
System (HIT), Harbin Institute of Technology, Harbin 150001, China (e-mail: performance in normal working conditions, the high tracking
huihuipan0@gmail.com; 1984sunweichao@gmail.com).
accuracy might be obtain by employing large feedback gains in
H. Gao is with the Research Institute of Intelligent Control and Sys-
tems, Harbin Institute of Technology, Harbin 150001, China (e-mail: practical implementations.
huijungao@gmail.com). In addition, noting that saturation nonlinearities occur in
X. Jing is with the Department of Mechanical Engineering, Hong Kong
many actuators, such as hydraulic servo valves, electric ser-
Polytechnic University, Hung Hom, Hong Kong. (e-mail: xingjian.jing@gmail.
com). vomotors, and other areas. In many engineering application
Digital Object Identifier 10.1109/TASE.2015.2414652 systems, the reliable operation and acceptable performance of

1545-5955 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING

most control systems must be assessed in light of actuator satu- to the plant, and is the output from the plant. rep-
ration [6], which may be possible to ignore this fact. However, resents the uncertain nonlinearities including external distur-
such unexpected actuator saturation may degrade the system bances. The actuator nonlinearity is described as a sat-
performance, render the instability of the closed-loop system. urated characteristic given by
Significant amount of analysis and synthesis of control systems
have been done to stabilize the system, while considering the (2)
saturation nonlinearities at the controller design stage, such as
the anti-windup techniques [7], [8] and the low-and-high gain where is the input saturation limit. In general, the system is
approaches [9], [10]. These designs focus on how to enlarge the subjected to structured uncertainties due to variations in system
domain of attraction by properly designing the linear feedback parameters. In addition, the partially known function ,
control laws while meeting certain performance requirements which represents the system unstructured uncertainty and the
such as disturbance rejection. However, in all those works, external disturbances that cannot be modeled by an explicit
it was assumed that the systems of concern are linear and function, is assumed to be a bounded constant. The control
exactly known, which, as stated earlier, is at variance with most objective of this paper is to design an adaptive controller
physical system in reality. for system (1) such that all signals in the closed-loop system
Therefore, it is necessary to concern with the problem of are bounded, while the output follows a desired reference
adaptive tracking control for a class uncertainty nonlinear signal with the tracking error converging to zero, where
system to follow a desired reference trajectory with actuator and its first derivatives are known and bounded. For
saturation, uncertainties and external disturbances. Meanwhile, the development of control laws, the following assumptions
by borrowing ideas from the ARC [5], [16] and to avoid em- are made.
ploying large feedback gains in practical implementations. An Assumption 1: In order to express convenient, we rewrite
approach that based on disturbance observer has been applied the function as a linear parameterized form:
for systems to compensate the strong effects of disturbances , where is a known contin-
[11]–[15]. Motivated by the observation, in this paper, a novel uous function vector and assumes that is
terminal sliding-mode-based nonlinear disturbance observer an unknown parameter vector which is to be estimated, are
(TSDO), where the modeling inaccuracy and disturbance are known constant vector values denoting the upper bound for its
integrated into a lumped disturbance, is designed to estimate absolute values.
the lump disturbances acting on the system. The proposed Remark 1: Although the unknown function can
TSDO can estimate the lumped disturbances and then used be represented in a linear parameterized form as above, a prac-
in the controller to compensate for them. This novel TSDO tical system is, in fact, mainly contain some uncertainty, and the
algorithm is designed to be equivalent to a low-pass filter, linear parameterized form is only its approximation. Thus, we
which can soften the switching control signal of the TSDO to can consider all possible uncertainties of the system as the parts
attenuate chattering. The effect of the filter can be properly . Therefore, it is important to estimate in real time
controlled by regulating the parameters in the algorithm. The for achieving good performances.
smooth control signal of the observer is directly used for the Assumption 2: Defining parameter set
parameter estimation in real time, and no phase lag of the , which represents the structured uncertainty, satisfies
estimation incurs. Then, integrating them via a robust control
action with some assumptions, a new saturation ARC with (3)
TSDO is proposed for a reference tracking system, such that
the output of the system achieve output tracking and meet the where and are
steady-state performance requirement in general. known, . In addition, we also assume that the
sign of is known, which implies that unknown constant is
II. PROBLEM FORMULATION strictly either positive or negative. Without losing generality,
we shall only consider the case when , namely,
This paper considers the same class of nonlinear dynamic sys- .
tems preceded by actuator saturation as in [17]–[19], which are
described by
III. CONTROLLER DESIGN

A. Parameter Adaptation
Let denote the estimate of and denote the estimation
error (i.e., ). The widely used projection mapping
(1) will be used to keep the parameter estimates within the
known bounded set as in [16] and [20]. Thus, a discontinuous
where , are some known continuous non-
projection is defined as
linear functions, is the system state vector defined by
assumed to be avail-
able, and parameters and represent unknown constants. (4)
is the output from the controller, is the actual input
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PAN et al.: DISTURBANCE OBSERVER-BASED ADAPTIVE TRACKING CONTROL WITH ACTUATOR SATURATION AND ITS APPLICATION 3

where . In (4), represents the th compoent of the where and are the nominal values of and , and
vector , and the operation for two vectors is performed in is a feedback gain.
terms of the corresponding elements of the vectors. Remark 2: If the exact system model is known and the actu-
Suppose that the parameter estimated is updated using the ator saturation is ignored, i.e., ,
following projection-type adaptation law: and , then the FL controller (10) can obtain
asymptotic tracking performance. However, in fact, all physical
(5) systems have modeling uncertainties. Hence,
and in general. From (10), one can note that the
where is a diagonal adaptation rate matrix and is an FL controller requires exact cancellation of nonlinearities to
adaptation function to be synthesized later. For any adaptation achieve the desired performance. In the presence of uncertain-
function , the projection law structure used in (5) guarantees ties, the nonlinearities may not get canceled exactly, which may
the following desirable properties hold [6]: result in poor performance, and thus, it is necessary to compen-
sate for the effects of the uncertainties.
(6) 2) Design of the Adaptive Robust Controller: To further im-
(7) prove performance, it is necessary to introduce different mech-
anisms to handle parametric uncertainty and unstructured un-
certainty separately. In this area, ARC proposed by Yao [5] is a
B. Controller Design good choice. If actuator saturation is ignored in the design, the
1) Design of the Feedback Linearization Controller: Let us ARC controller can handle both parametric uncertainty by adap-
consider the following virtual control effort and the tracking tive control and unstructure uncertainty via deterministic robust
error in th step feedback control and integrate these two control techniques in
a unified form by using a projection-based adaptive law. Based
on (9), the resulting ARC controller can be given by
(8)

where is the output tracking error and is the virtual

control at the th step to be determined.
a) Step i : We select the virtual control law
as , (11)
where is positive real number feedback gain.
The scalar with and By substituting (11) into (9), we have

being integers represents the addition of all the terms (12)

corresponding to all the possible combinations of non-negative
where is the regressor for parameter adaptation.
integers, , satisfying . For
Noting Assumption 2 and P1 in (6), the robust function
example, when and
is then chosen to satisfy the following robust performance con-
ditions [20]:
.
b) Step n. From (1) and (8), the error dynamics can be ob- (13)
tained as
where is a positive design parameter that can be arbitrarily
small. Then, one smooth example of satisfying (13) is given
by

(14)

(9) where can be regarded as a nonlinear feedback gain, is any

smooth function satisfying , where
Since and is the upper bound of .
is a stable transfer function, making small or converging to Lemma 1: Consider the parameter estimates updated by the
zero is equivalent to making small or converging to zero. projection-type adaptation law (4), and the adaptation function
Thus, the rest of the design is to make as small as possible. in (5) is chosen as and defining , thus, the
Then, based on the system model, a FL controller is designed ARC law (11) guarantees that all signals are bounded. Further-
as follows: more, the positive-definite function defined by is
bounded by .
Proof: The proof procedure can use the same techniques as
in [5]. Therefore, it is easy to ensure that the results in Lemma
(10) 1 are true.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING

It seems that the control problem with uncertainties has been the sliding-mode surface in finite-time. In order to derive the
solved by the ARC controller (11). However, this controller TSDO algorithm for system (16), a sliding function is defined
might not work very well in practice. Considering the robust as follows:
control law (14) and condition (13) carefully, it seems that the
robust control law acts as a high-gain feedback law, which uses (17)
the maximum bound of uncertainties. In fact, it is difficult to
implement high-gain feedback control in practice due to mea- where and are constants. is designed
surement noise and unmodeled dynamics. As a result, high-gain such that the polynomial , which
feedback control may lead to actual system being unstable. To corresponds to system (17), is Hurwitz, i.e., the eigenvalues of
avoid this practical issue, finite gain and large have to the polynomial are all in the left-half side of the complex plane.
be employed. These parameter choices will make the results are positive values and satisfy the following condition as
of Lemma 1 have little sense. Hence, when unstructured un- [21]:
certainty is dominate in physical systems, model-based adap-
tive control can do little to handle it, and limited feedback gains (18)
cannot achieve excellent tracking performance anymore.
3) Design of the Adaptive Robust Controller With TSDO: where .
In order to dominate the disturbances coming from various un- When the state trajectories of the error system (16) enter into
certainties without high-gain feedback, a TSDO is designed to the ideal sliding-mode , the nonlinear system (17) will
compensate the uncertainties and to attenuate the external dis- behave in an identical fashion, namely
turbances. For the purpose of uncertain parameters estimation,
considering the system dynamic model (1), a sliding-mode ob-
server can be designed as follows:
(19)
(15) If in sliding-mode manifold (17) are selected using (18) and
in (17) are determined to guarantee that the polynomial
where represents state estimation for , which corresponds to system (17),
is the error compensation signal of the is Hurwitz, system (19), which represents the establishment of
TSDO, which will take the feedback into the design of the the ideal sliding-mode for system (17), can converge to
observer; and and are two nominal parameters that are its equilibrium point from any initial condition
some crude estimations of the true parameters in (1). and along the sliding-mode manifold in finite-time
can be determined initially based on the experiences and [21]–[23].
prior knowledge. It should be noticed that accuracy is not Assumption 3: The derivative of in system
necessary for either parameter. Although the exact values of (16) is bounded: , where
the system parameters and their estimations are unknown, one is a constant.
can reasonable assume that and , i.e., the errors between Theorem 2: For the system described by (1) with an estima-
the true system parameters and their crude estimations, are tion model (15), the tracking error systems defined by (16) sat-
bounded. isfy that the error dynamics converge to zero along within
Remark 3: In fact, it is not necessary to estimate the state finite-time, if the sliding-mode surface is chosen as (17) and
using (15). All of them are assumed to be directly measured the control laws are given by
using sensors. The purpose of the observer (15) is for the esti-
mation of the system errors parameters and and uncertain (20)
parameter . (21)
Defining the tracking errors as (22)
, then subtracting (1) from (15) leads to error (23)
dynamics systems as where repre-
sents the lump disturbances. The error compensation signal where and are all constants, as
is an control signal whose design will be described later. defined in (17); is a positive constant; is a constant defined
Now, the task of TSDO for nonlinear errors system (16), in Assumption 3; two constants, and are selected to
shown at the bottom of the page, is to design a control strategy satisfy the following condition:
which induces an ideal sliding-mode motion in the prescribed
sliding-mode surface and forces system (16) to the origin along (24)

(16)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PAN et al.: DISTURBANCE OBSERVER-BASED ADAPTIVE TRACKING CONTROL WITH ACTUATOR SATURATION AND ITS APPLICATION 5

Proof: From system (16), the sliding-mode manifold (17) or decreases. It should be noted that to obtain is much
can be rewritten as follows: easier than to obtain the accurate value of .
Remark 5: As stated in Theorem 2, the control signal (22) is
equivalent to a low-pass filter, where is the input and
is the output of the filter. The transfer function corresponding
to the filter (22) is , where the
Substituting the control (20) and (21) into above equation gives
bandwidth of the low-pass filter is equal to the parameter .
(25) Although the control law in (23) is discontinuity caused
by the switch function in (20) is the output of the
If the sliding-mode manifold , then it is not difficult to low-pass filter (22) and is softened to be a smooth signal by
get , that is, (22). From the proof of Theorem 2, it can be seen that the value
. On both sides of the inequality is multiplied by , one has of in the low-pass filter does not affect the stability of
. From (24), the following relationships under the observer. Therefore, can be designed arbitrarily based on
the condition can be obtained: , and the requirement of softening , given that the condition of
. (24) is satisfied. Hence, the control signal of the observer, ,
Similarly, when the sliding-mode manifold , then we cannot induce the chattering phenomenon in the system because
have . Finally, once the sliding-mode man- of the low-pass filter and can be directly used for the parameter
ifold , one has . estimation.
The aforementioned analysis yields to the following in- For system (1), its observer (15) can be designed according
equality kept forever: to Theorem 2. The error system (16) will converge to
in finite time. When , from (25), it can be obtained as
(26)
. Therefore, the estimation of
Consider the following Lyapunov function: . For can be obtained as follows:
sliding-mode manifold (17), its derivative with respect to time
(30)
along system (16) can be obtained from (25) as
. Substituting (22) into above equa- Remark 6: From (30), it can be seen that can be
tion yields estimated in real time because of the smooth . The gain of
(27) the switch function in (23) can be designed for guaranteeing
the robustness, while the equivalent low-pass filter in (22) can
Therefore, it is easy to obtain that soften and produce a smooth signal , which can be
directly used for estimating .
Now, the lump disturbance has already been perfect esti-
mated. In the last step (step ) of ARC, let the control law be
(28)
(31)
From (26), Assumption 3 and above equation, thus, the time-
derivative of the Lyapunov function candidate along system tra- where
jectory can be rewritten as
(32)
(29)

which means that system (16) will reach to in finite time.

On , system (16) behaves in an identical fashion, as shown
in (19), i.e., the system will converge to zero in finite-time along
. This completes the proof. (33)
In fact the proposed control provides for the establishment of
the th-order sliding-mode in (34)
the extended state space with coordinates .
Remark 4: It should be mentioned that in (20)–(23), all vari- where is a positive constant.
ables are available except on account of is not available in In the above, the expression of control is divided into three
(16). For calculating in (23), the function has to be regions. Region I with represents the linear operation
considered range of the system with is chosen to meet the performance
can be calculated through the requirement at the steady-state. Region III with uses
equation , where is a time the maximum control input to prevent from going too large.
delay constant and can be chosen as a fundamental sample time Region II is a smooth transition region between region I and
because . In this way, region III.
the accurate value of needn't to be known except its sign, We are interested in finding the positive invariant set of the
, namely, we only need to known whether increases tracking error . Specifically, once the tracking errors are within
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING

this positive invariant set, they will not go out of this set there- Theorem 4: Considering the lump uncertainty and
after. And the control effort needed to confine the error within its estimation (30), and letting the parameter estimation be up-
the positive invariant set is always within the saturation limits dated by the adaptation law (5) in which is chosen as Lemma
so that no saturation would occur. Such a set is given below. (1), then the control law (31) guarantees that all signals are
Lemma 3: With the control law (31), the set bounded and converge to in the presence of actuator satura-
is a positive invariant set and if tion. Furthermore, the proposed control law TMDO-ARC (31)
the following conditions are satisfied: guarantees that, after the finite time , the lump uncertainty is
1) ; well estimated, (i.e., ), then, zero final output
2) ; tracking error with respect to the desired trajectory is achieved,
3) i.e., as .
. Proof: By substituting (33) into (9), we have
Proof: Involving 2) and 3), it is obtained that

(35)

For , from (29), we can concluded that .

By integrating the aforementioned inequality from 0 to , we
By comparing the above inequality with the control input law
have , which leads that , i.e.,
(31), it can be easily shown that . Thus,
. Therefore, from (35) we can
inside , the dynamics of can be simplified into
see that is bounded for .
For , from the fact (30), the derivative of along with
(35) is given by

where
.
Suppose that . Then, some ele- Since and is bounded, signal and
ments must hit their boundaries or while others are still are thus uniformly continuous with respect to time.
within their limits. Let denote the elements Then, it can be verified that signal
that hit their upper bounds, i.e., . Let is also uniformly continuous. Furthermore, since
denote the elements that hit their lower bounds, i.e., . and d
Let denote the elements that are within their is lower bounded. By applying Barbalat's lemma,
limits, i.e., . Then it is easy to see that, the as . Therefore, we can
vector always points inward if the following conditions are obtain that as .
satisfied: and . Remark 7: It notes that the results in Theorem 4 are better
These conditions are verified as follows. than that in Lemma 1. Particularly, the proposed control is easier
1) If , then, we have . Based on condition 3) to be implemented and high-gain feedback is also avoided. This
in Lemma 3, which further implies that controller indeed has better performance which will be illus-
, which tells us trated in the following comparative experiments.
that when hits the upper bound will be decreased, IV. APPLICATION TO A QUARTER VEHICLE
and back to the set of . SUSPENSION SYSTEM
2) Conversely, if , which means hits the lower
In this section, the proposed control algorithm is applied to an
bound , then
quarter active suspension controller. Consider the experimental
, which means that when hits the lower
example of a bench-scale quarter-car model with active suspen-
bound will be increased, and back to the set of
sion system in [24] and [25]
as well.
3) If , which is easy to verify that
.
(36)
4) Similarly, if , we can obtain that
. where the sprung mass represents the body of the car; the
This completes the proof that is a positive invariant. unsprung mass denotes the wheels, brakes and part of the
suspension; and denote the forces produced by the springs
C. Main Result and dampers, and are the elasticity force and damping
In this section, by using the disturbance observer in the afore- force of the tires. and are the displacements of the sprung
mentioned section, a saturated adaptive robust control is de- and unsprung masses, respectively; is the road displacement
veloped, and the stability analysis of the closed-loop system is input; represents the saturated control input. It is to be
given. pointed that with a change in the number of passengers or the
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PAN et al.: DISTURBANCE OBSERVER-BASED ADAPTIVE TRACKING CONTROL WITH ACTUATOR SATURATION AND ITS APPLICATION 7

payload, the vehicle load will easily vary and this will accord-
ingly change the vehicle mass . Thus, is an uncertain pa-
rameter, which extent is known, i.e., ,
where kg and kg.
The forces produced by the nonlinear suspension stiffness
spring, and the piecewise linear damper and the tire obey

(37)

(38)

(39)

where and are the stiffness coefficients of linear and non-

linear terms; and are the damping coefficients for the ex-
tension and compression movements; and are the stiff-
ness and damping coefficient of the tire, respectively. The sig-
nificance and values of the active suspension parameters under
consideration are referred to [25]. To implement the experiment,
we assume the unknown mass kg, and the actuator
saturation limit N. Fig. 1. Vertical displacement.
By appropriate coordinate transformation as [24], the active
suspension model (36) can be transferred to two subsystems,
including a two-order sprung one and unsprung one, where the
sprung one satisfies the formation of (1). The proposed method
is employed for the sprung subsystem of the active suspension
system. In addition, to illustrate the effectiveness of the pro-
posed method (C4), the passive suspension (C1), FL controller
(C2), and standard ARC (C3) are employed for performance
comparison. The detail control design is omitted here for space
due to its simplicity.
Since the switch parameter in observer can
be regarded as one parameter, thus, only gain needs to be
tuned during operating. The controller parameters are designed
as
. The
feedback linearization controller and adaptive robust controller
are chosen the same parameters as the corresponding parameters
in the proposed controller.
The suspension system is evaluated by subjecting to a 4 Hz
sine wave of 0.3 cm amplitude road disturbance, that is, Fig. 2. Control input.
. Response curves of the state variables under the
four methods are shown in Fig. 1. The corresponding control
signals are shown in Fig. 2. 59.01%, respectively. It is not hard to see that the improvement
From the experiment results in Fig. 1, it can be seen that the in terms of quality of control achieved by the TSDO-ARC al-
active suspension induces an important improvement of ride gorithm is the best. These improvements are significant.
performance compared to passive one. The responses using the
proposed algorithm C4 show an improvement with respect to V. CONCLUDING REMARKS
C2 and C3. Finally, to further evaluate the suspension system This paper has presented a new design scheme for saturated
performance, the root mean square (RMS) values of the ver- adaptive robust tracking controllers of a class uncertain non-
tical displacement are exploited to demonstrate the effective- linear systems subjection to modeling uncertainties, actuator
ness of the proposed control design method. RMS values are saturation, and bounded extern disturbance. Based on the
strictly related to the ride comfort, which are often employed TSDO, the proposed TSDO-ARC guarantees the finite time
to quantify the amount of disturbance transmitted to the vehicle of estimation error to a time-varying disturbance. The global
body. The steady-state body vertical displacement values cor- asymptotic stability of the combined controller-observer system
responding to C1, C2, C3, and C4 are, respectively, equal to is shown using the Lyapunov analysis. Simulation studies
0.0015, 0.0012, 0.0011, and 0.0006148. Consequently, the im- were conducted on the track-following of a suspension system.
provements in terms of quality of control that were achieved The results have demonstrated the advantages of the proposed
by the three control algorithm are equal to 20%, 26.67%, and TSDO-ARC over the conventional FL and ARC controller.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING

REFERENCES Huihui Pan received the M.S. degree in control

[1] D. Xu, B. Jiang, and P. Shi, “Adaptive observer based data-driven con- science and engineering from the Harbin Institute of
trol for nonlinear discrete-time processes,” IEEE Trans. Autom. Sci. Technology, Harbin, China, in 2013. He is currently
Eng., vol. 11, no. 4, pp. 1037–1045, Oct. 2014. working toward the Ph.D. degree at the Research
[2] Q. Wei and D. Liu, “Adaptive dynamic programming for optimal Institute of Intelligent Control and Systems, Harbin
tracking control of unknown nonlinear systems with application to Institute of Technology.
coal gasification,” IEEE Trans. Autom. Sci. Eng., vol. 11, no. 4, pp. His research focuses mainly on vehicle suspension
1020–1036, Oct. 2014. control, adaptive control, sliding-mode control and
[3] K. Pathak, J. Franch, and S. Agrawal, “Velocity and position control of robust control.
a wheeled inverted pendulum by partial feedback linearization,” IEEE
Trans. Robot., vol. 21, no. 3, pp. 505–513, Jun. 2005.
[4] S. Talole, J. Kolhe, and S. Phadke, “Extended-state-observer-based
control of flexible-joint system with experimental validation,” IEEE
Trans. Ind. Electron., vol. 57, no. 4, pp. 1411–1419, Apr. 2010.
[5] L. Lu and B. Yao, “A performance oriented multi-loop constrained
adaptive robust tracking control of one-degree-of-freedom mechan- Weichao Sun (S’13) received the Ph.D. degree in
ical systems: Theory and experiments,” Automatica, vol. 50, no. 4, pp. control science and engineering from the Harbin In-
1143–1150, 2014. stitute of Technology, Harbin, China, in 2013.
[6] Y. Hong and B. Yao, “A globally stable saturated desired compensa- Since 2013, he has been with the Research Institute
tion adaptive robust control for linear motor systems with comparative of Intelligent Control and Systems, Harbin Institute
experiments,” Automatica, vol. 43, no. 10, pp. 1840–1848, 2007. of Technology.
[7] A. Teel and N. Kapoor, “The L2 anti-windup problem: Its definition
and solution,” in Proc. Eur. Control Conf., 1997, pp. 1–4.
[8] G. Grimm, J. Hatfield, I. Postlethwaite, and A. Teel, “Antiwindup for
stable linear systems with input saturation: An LMI-based synthesis,”
IEEE Trans. Autom. Control, vol. 18, no. 9, pp. 1509–1525, Sep. 2003.
[9] T. Hu and Z. Lin, Control Systems With Actuator Saturation: Analysis
and Design. New York, NY, USA: Springer, 2001.
[10] A. Saberi, Z. Lin, and A. Teel, “Control of linear systems with sat-
urating actuators,” IEEE Trans. Autom. Control, vol. 41, no. 3, pp.
368–378, Mar. 1996. Huijun Gao (SM’09–F’14) received the Ph.D.
[11] A. Bartoszewicz and P. Lesniewski, “Reaching law approach to the degree in control science and engineering from the
sliding mode control of periodic review inventory systems,” IEEE Harbin Institute of Technology, Harbin, China, in
Trans. Autom. Sci. Eng., vol. 11, no. 3, pp. 810–817, Jul. 2014. 2005.
[12] J. Yang, S. Li, and X. Yu, “Sliding-mode control for systems with mis- From 2005 to 2007, he carried out his postdoc-
matched uncertainties via a disturbance observer,” IEEE Trans. Ind. toral research with the Department of Electrical
Electron., vol. 60, no. 1, pp. 160–169, Jan. 2013. and Computer Engineering, University of Alberta,
[13] P. Ignaciuk and A. Bartoszewicz, “Sliding mode dead-beat control of Canada. Since November 2004, he has been with the
perishable inventory systems with multiple suppliers,” IEEE Trans. Harbin Institute of Technology, where he is currently
Autom. Sci. Eng., vol. 9, no. 2, pp. 418–423, Apr. 2012. a Professor and Director of the Research Institute
[14] X. Zhao, H. Liu, J. Zhang, and H. Li, “Multiple-mode observer design of Intelligent Control and Systems. His research
for a class of switched linear systems,” IEEE Trans. Autom. Sci. Eng., interests include network-based control, vehicle dynamic control, robust
vol. 12, no. 1, pp. 272–280, Jan., 2015. control/filter theory, time-delay systems and their engineering applications.
[15] Y. Feng, X. Yu, and F. Han, “High-order terminal sliding-mode ob- Dr. Gao is the recipient of the IEEE J. David Irwin Early Career Award
server for parameter estimation of a permanent-magnet synchronous from IEEE IES. He is a Co-Editor-in-Chief for the IEEE TRANSACTIONS
motor,” IEEE Trans. Ind. Electron., vol. 60, no. 10, pp. 4272–4280, ON INDUSTRIAL ELECTRONICS, and an Associate Editor for Automatica, the
Oct. 2013. IEEE TRANSACTIONS ON CYBERNETICS, the IEEE TRANSACTIONS ON FUZZY
[16] J. Yao, Z. Jiao, D. Ma, and L. Yan, “High-accuracy tracking control of SYSTEMS, the IEEE/ASME TRANSACTIONS ON MECHATRONICS, and the IEEE
hydraulic rotary actuators with modeling uncertainties,” IEEE Trans. TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY. He is serving on the
Mechatronics, vol. 19, no. 2, pp. 633–641, Apr. 2014. Administrative Committee of the IEEE Industrial Electronics Society (IES).
[17] W. Gao and R. Selmic, “Neural network control of a class of nonlinear
systems with actuator saturation,” IEEE Trans. Neural Netw., vol. 17,
no. 1, pp. 147–156, Jan. 2006.
[18] Z. Zhang, S. Xu, and B. Zhang, “Asymptotic tracking control of un-
certain nonlinear systems with unknown actuator nonlinearity,” IEEE
Trans. Autom. Control, vol. 59, no. 5, pp. 1336–1341, May 2014.
[19] W. Sun, Z. Zhao, and H. Gao, “Saturated adaptive robust control for Xingjian Jing (M’13) received the B.S. degree
active suspension systems,” IEEE Trans. Ind. Electron., vol. 60, no. 9, from Zhejiang University, Hangzhou, China, in
pp. 3889–3896, Sep. 2013. 1998, the M.S. degree in intelligent systems and
[20] J. Yao, Z. Jiao, and D. Ma, “Adaptive robust control of DC motors with control, and the Ph.D. degree in mechatronics from
extended state observer,” IEEE Trans. Ind. Electron., vol. 61, no. 7, pp. the Shenyang Institute of Automation, Chinese
3630–3637, Jul. 2014. Academy of Sciences, Shenyang, China, in 2001 and
[21] S. Bhat and D. Bernstein, “Geometric homogeneity with applications 2005, respectively, and the Ph.D. degree in nonlinear
to finite-time stability,” Math. Control, Signals. Syst., vol. 17, no. 2, pp. systems and signal processing from the Department
101–127, 2005. of Automatic Control and Systems Engineering,
[22] Y. Feng, F. Han, and X. Yu, “Chattering free full-order sliding-mode University of Sheffield, Sheffield, U.K., in 2008.
control,” Automatica, vol. 50, no. 4, pp. 1310–1314, 2014. He is currently an Assistant Professor with the
[23] Y. Hong, Y. Xu, and J. Huang, “Finite-time control for robot manipu- Department of Mechanical Engineering, Hong Kong Polytechnic University
lators,” Syst. Control Lett., vol. 46, no. 4, pp. 243–253, 2002. (PolyU). His current research interests include: nonlinear analysis and design
[24] W. Sun, H. Pan, Y. Zhang, and H. Gao, “Multi-objective control for in the frequency domain, system identification, signal processing and control
uncertain nonlinear active suspension systems,” Mechatronics, vol. 24, of complex nonlinear systems, intelligent computing methods, and their
no. 4, pp. 318–327, 2014. applications in nonlinear mechanical systems (sound and vibration control),
[25] W. Sun, H. Gao, and O. Kaynak, “Vibration isolation for active sus- nonlinear physiological systems (neural systems) and robotic systems, etc.
pensions with performance constraints and actuator saturation,” IEEE/
ASME Transa. Mechatronics, vol. 20, no. 2, pp. 675–683, Apr. 2015.