ISA Transactions 53 (2014) 671–678

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Practice Article

A novel disturbance-observer based friction compensation scheme

for ball and plate system
Yongkun Wang a, Mingwei Sun a,n, Zenghui Wang b, Zhongxin Liu a, Zengqiang Chen a
a
Department of Automation, Nankai University, Nankai District, Tianjin 300071, China
b
Department of Electrical and Mining Engineering, University of South Africa, Florida 1710, South Africa

art ic l e i nf o a b s t r a c t

Article history: Friction is often ignored when designing a controller for the ball and plate system, which can lead to
Received 30 November 2012 steady-error and stick-slip phenomena, especially for the small amplitude command. It is difficult to
Received in revised form achieve high-precision control performance for the ball and plate system because of its friction. A novel
17 October 2013
reference compensation strategy is presented to attenuate the aftereffects caused by the friction. To
Accepted 12 November 2013
realize this strategy, a linear control law is proposed based on a reduced-order observer. Neither the
Available online 14 December 2013
This paper was recommended for accurate friction model nor the estimation of specific characteristic parameters is needed in this design.
publication by Prof. Yang Quan Chen Moreover, the describing function method illustrates that the limit cycle can be avoided. Finally, the
comparative mathematical simulations and the practical experiments are used to validate the effective-
Keywords: ness of the proposed method.
Ball and plate system
& 2013 ISA. Published by Elsevier Ltd. All rights reserved.
Friction compensation
Disturbance observer
Active disturbance rejection control
Limit cycle

1. Introduction investigated. In [4], a control method using switching-driving

The ball and plate system, a two-dimensional extension of the
ball and beam system [1], has strong coupling and nonlinear
dynamics [2]. It is a highly unstable plant with apparent friction
effects, which is popularly used as a benchmark to test the
effectiveness of specific control algorithms, especially in the field
of friction compensation. The plate with two rotating axes is the
control target. A digital camera is placed above the plate to
measure the relative position of a copper ball with regard to the
center of the plate. The control objective, by rotating the plate, is to
stabilize the ball on a specific position or make it go along a certain
trajectory.
Up to now, many control methods have been proposed and
used to control the ball and plate systems. PID control algorithm is
the most widely used control method for the ball and plate
system. However it is difficult for using the traditional PID
controller to obtain excellent control performance. The reference
[3] demonstrated that the PID controller has a lot of limitations for
the ball and plate system, such as unavoidable overshoot and long
settling time. To achieve good control performance for the ball and
plate system, several modern control strategies have been
n
Corresponding author. Tel.: þ 86 13821839352.
E-mail addresses: wangyongkun1029@126.com (Y. Wang),
smw_sunmingwei@163.com, sun_mingwei@aliyun.com (M. Sun),
wangzengh@gmail.com (Z. Wang), lzhx@nankai.edu.cn (Z. Liu),
chenzq@nankai.edu.cn (Z. Chen).
Lyapunov function was proposed, but this control algorithm is
too complicated to be applied in practice. Two fuzzy control
methods were presented in [5,6]. However, the method proposed
in [5] was only tested by the mathematical simulation; and the
steady-error is unable to be eliminated for the method in [6]. All
these proposed control methods were designed without consider-
ing the friction, which could limit the performance to be further
improved.
The friction in mechanical systems is a nonlinear phenomenon
that has diverse aspects giving rise to control problems such as
steady-error, limit cycle, stick-slip etc, which cause severe perfor-
mance degradation [7]. Many friction compensation algorithms
[8–11] have been proposed based on the friction mechanism
model in terms of velocity information. The essence of traditional
compensations is to estimate the crucial parameters of the friction
model. However the high-precision control performance is still
difficult to be achieved in the low-velocity regime when imple-
menting these traditional algorithms in practice. The major reason
originates from the discontinuity of Coulomb friction at velocity
reversals. Thus, the compensation term switches its sign fre-
quently during the steady-state phase, causing chattering phe-
nomenon. In addition, overcompensation or undercompensation is
inevitable for the friction model based direct compensation
approaches. Undercompensation of friction could lead to steady-
error and overcompensation may induce limit cycles [12,13]. In
conclusion, it is necessary to develop a new type of friction
compensation scheme which can solve the mentioned problems.

0019-0578/$ - see front matter & 2013 ISA. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.isatra.2013.11.011
672
Han proposed a novel philosophy—Active Disturbance Rejec-
tion Control (ADRC) algorithm based on the feedback linearization
philosophy, which does not rely on a refined dynamic model
[14,15]. ADRC is a disturbance observer based control method and
has demonstrated its effectiveness in many fields [16–23]. In this
paper, ADRC strategy is utilized to reject the friction effect at low
velocity. A reduced-order disturbance observer is proposed to
attenuate the aftereffects caused by the friction on the output of
the traditional PD-type control scheme instead of compensating
the primary friction mechanics. In other words, a pseudo reference
command is produced to counteract the output error during the
steady-state phase. Moreover, the describing function technique is
used to test whether the proposed method can generate limit
cycles or not.
The rest parts of the paper are organized as follows: Section 2
formulates the problem. Section 3 derives the disturbance obser-
ver based linear control laws for the reference compensation
strategy. In Section 4, the analysis of limit cycle is performed
using the describing function technique. Section 5 provides the
comparative numerical simulations and experiments to verify the
proposed methodology. Finally, the concluding remarks are given
in the last section.
2. Problem formulation
On the basis of the Lagrange method, the nonlinear equations
of the ball and plate system can be formulated as [2]
m Based on the mature MSMD O121P1U board, the original
x€ ¼ _  mg ð sin α μ cos αÞ;
½xα_ 2 þ α_ βy
m þ I b =r 2 m þ I b =r 2
m which can be represented by a first-order model
y€ ¼
2
_  mg ð sin β  μ cos βÞ;
½yβ_ þ α_ βx
m þ I b =r 2 m þ I b =r 2
1 1
α€ ¼  ð2mxx_ α_ þ mxyβ€ þ mx_ yβ_ þ mxy_ β_ þmgx cos αÞ þ
ðI p þI b þmx2 Þ
and
β€ ¼ 
1
ð2myy_ β_ þ mxyα€ þmx_ yα_ þ mxy_ α_ þmgy cos βÞ þ
ðIp þ Ib þ my2 Þ
where m and r are the mass and radius of the ball respectively, g is
the gravitational constant, Ib is the mass moment of inertia of the
ball, Ip is the mass moment of inertia of the plate, x and y are the
relative positions of the ball on the two axes of the plate with
regard to the center respectively, x_ and y_ are the corresponding
velocities, α and β are the angles of the plate in x-axis and y-axis
respectively, μ is the sliding friction coefficient between the ball
and the plate, τx and τy are the torques applied to the plate in the
x-axis and y-axis respectively. It should be noted that the friction
_ are included
torques with respect to the angular velocities, α_ and β,
in τx and τy, respectively.
In practice, the x axe and the y axe are weakly coupled for a
static position control [5]. Hence, the control laws for these two
axes can be designed separately. The block diagram for each axis is
shown in Fig. 1. Here, the cascade control structure is employed,
where the angular reference produced by the outer-loop position-
error based controller is fed into the inner-loop controller to track.
The inner-loop system is an angular servo system and is driven by
MSMD O121P1U board developed by Panasonic Corporation. In the
experiment, a fixed angular PD controller is employed on the
board, which is unable to be modified. For this system, there is no
way to measure the angular information. When the system is in
operation, the steady-error and the limit cycles will emerge
when the PD controller and the PID controller are employed in
Y. Wang et al. / ISA Transactions 53 (2014) 671–678
the outer-loop position control, respectively. These problems are
clear for small amplitude position commands. As mentioned in the
introduction, these problems are caused by the friction of the
system. Hence, an efficient compensation method is desired to
compensate the friction and obtain high-precision control
performance.
3. Disturbance observer based compensation strategies
In this section, two compensation methods are presented based
on disturbance observers. Firstly, we give the design process of a
reference compensation strategy. And then, a direct compensation
strategy is described. It means that an observer is added to the
position PD controller in the original control structure. Without
loss of generality, this paper only gives the control design process
for the x-axis.
3.1. Reference compensation strategy
For the system studied in this paper, the inner-loop angular
control is fixed and the friction model based direct compensation
cannot be realized. However, with respect to this practical situa-
tion, it is possible that a pseudo-position reference, which is fed
into the outer position loop, can be designed to achieve zero
steady-error. The reference compensation strategy is shown in
Fig. 2. This is a triple loop cascade system with the two inner loops
unchanged.
ð1Þ position control system is well-tuned by the developer and the
system exhibits over-damped characteristic without limit cycle,
ð2Þ
xðsÞ Km
GP ðsÞ ¼ ¼ ð5Þ
rðsÞ τs þ 1
τx
ðI p þ I b þmx2 Þ where r is the pseudo-reference command as shown in Fig. 2. The
ð3Þ Eq. (5) can be reformulated as
τx_ þ x ¼ K m r; ð6Þ
1
τy
which can be described in a state-space form
ðI p þ I b þmy2 Þ 8
_
ð4Þ < x1 ¼ x2 þ br;
>
x_ 2 ¼ w; : ð7Þ
>
:y ¼x ;
o 1
where
8
< x1 ¼ x;
>
x2 ¼  x=τ; : ð8Þ
>
: b ¼ K =τ;
m
where yo is the measurable output, x2 is known as an extended
state [14] and w is its derivative. In x2, the consequences caused by
the friction torque, the disturbance torque and the model uncer-
tainties can be assembled. It is not a real physical state but a
combined signal. w is a pseudo manipulation variable that gen-
erates x2 and terminates the description of the states. It should be
noted that any discontinuous property of the velocity, such as the
Coulomb friction, can be transformed via a physical integrator
naturally contained in the motor into a continuous format at the
angular position end. This establishes a favorable condition for
employing conventional estimation schemes.
Now, we present the estimation method for obtaining x2 to be
used in the subsequent control design. Since the state x1 can be
measured directly, a reduced-order observer is introduced to
decrease the phase lag as
z_ 2 ¼ lð  br  z2 þ x_ Þ ð9Þ
 Y. Wang et al. / ISA Transactions 53 (2014) 671–678 673

Fig. 1. Block diagram of control configuration for ball and plate system.

Fig. 2. Block diagram of the reference compensation scheme.

where z2 is the estimation for x2 and l is the observer gain. As x_ is the friction, which leads to the position steady-error, can be
not available, we define a new state eliminated too.
zc ¼ z2  lx ð10Þ
which can yield a linear reduced-order state observer (11) for (7) 3.2. Direct compensation strategy
z_ c ¼  lz2  lbr ð11Þ
Here, a direct position compensation strategy based on a
Because x2 is the extended state, the observer, i.e. the combination disturbance observer is presented to make further comparisons.
of (10) (z2 ¼ zc þlx) and (11), is called Extended State Observer Without changing the original position PD controller, a second-
(ESO) [14]. The observer gain l can be denoted as ωo, the observer order extended state observer is added to compensate the error
bandwidth. With a well-tuned observer, one can obtain z2 Ex2. If effects caused by the friction. This is a double loop cascade control
the pseudo-reference signal is described in the form of structure as shown in Fig. 1 where the compensation is supple-
r ¼ ðr 0 z2 Þ=b ð12Þ mented to the outer position control loop.
As the angular loop has fast and accurate tracking performance,
where r0 is a virtual control variable, the original system (6) can be
it can be approximately considered as a unit gain when designing
reformulated as
the compensator, which means that αEαr and the dynamic plant
x_ ¼ r 0 ð13Þ from the angular command αr to the position xd can be described
by Eq. (1). Because the sliding friction coefficient μE0 and
In a word, the essential function of the observer is to simplify the
assuming that the plate rotates within small angles, Eq. (1) can
original system (5) to a pure integrator where the new control
be approximated by
variable is r0. This is just the philosophy of feedback linearization
however in the absence of a refined model. It is reasonable to m
x€ d ¼ _  mg α;
½x α_ 2 þ α_ βy ð16Þ
employ the classical PD control form for r0: m þ I b =r 2 d m þ I b =r 2
r 0 ¼ ke ðxc  xÞ kv x_ ð14Þ
whose state-space representation is
where xc is the desired position. In summary, the reference
(
compensation becomes x€ d ¼ f d þ bd α;
ð17Þ
r ¼ ðke ðxc  xÞ  kv x_ z2 Þ=b ð15Þ yd ¼ xd ;

where ke and kv are the gains of the virtual control variable. Now
the design process is finalized for the reference compensation where yd is the measurement output, and
strategy. It can be seen that the controller is in a complete linear 8
<fd ¼ m _
½x α_ 2 þ α_ βy;
form, so it is called Linear Active Disturbance Rejection Control m þ I b =r 2 d
ð18Þ
(LADRC) [24]. Although the LADRC is linear-time-invariant, it can : bd ¼  mg=ðm þ I b =r 2 Þ:
be applied to control the nonlinear, time-variant and uncertain
processes with very little model information. Define xd,1 ¼yd, xd;2 ¼ x_ d;1 and xd,3 ¼fd (extended state), Eq. (17) can
Remark 1. The friction can cause position steady-error for Fig. 1. It be rewritten as
can be thought that there is a disturbance in the control output (r) 8
> x_ ¼ xd;2 ;
or the position command, which produces steady-error. This < d;1
disturbance is also included in the extended state, i.e. the total x_ d;2 ¼ xd;3 þ bd α; ð19Þ
>
: x_ ¼ w ;
disturbance, which makes the plant deviate from an ideal inte- d;3 d
grator. The total disturbance can be captured by the observer
quickly, and then an opposite control action can be implemented where wd is assumed to be the source of the total disturbance. As
to cancel it. In such a way, the equivalent disturbance triggered by aforementioned, the reduced-order observer for (19) can be
674 Y. Wang et al. / ISA Transactions 53 (2014) 671–678

chosen as armature respectively. The linearized transfer function of Ga is

" #  " z #   " # " #!
z_ d;2 0 1 d;2 bd l1   zd;2 αðsÞ Kt
¼ þ αþ y_ d  1 0 Ga ðsÞ ¼ ¼ ; ð26Þ
z_ d;3 0 0 zd;3 0 l2 zd;3 uðsÞ JLs3 þ JRs2 þ K t K e s þ Nð U ÞsðR þ LsÞ

ð20Þ where u is the control voltage.

T T T The reference compensator (15) is equivalent to a PID controller
where [zd,2,zd,3] is the estimation for [xd,2,xd,3] , and L ¼[l1,l2] is
CPID with regard to the stability margin and the corresponding
the observer gain vector. Proceeding along the same line as in
P–I–D gains are
Section 3.1, the new states can be defined as 8
" # " # " # < K p ¼ ðke þ ωo kv þ ωo Þ=b;
>
zc;2 zd;2 l1
¼  y ð21Þ K i ¼ ωo ke =b; ð27Þ
zc;3 zd;3 l2 d >
: K ¼ k =b:
d v

and the reduced-order observer for (19) is formulated as

8 According to Fig. 2, specifying the closed-loop characteristic
>
> z_ c;2 ¼  l1 zd;2 þ zd;3 þ bd α;
>
>
< z_ c;3 ¼  l2 zd;2 ;
>
> zd;2 ¼ zc;2 þ l1 xd ;
>
> ð28Þ
: zd;3 ¼ zc;3 þ l2 xd ;
where [zc,2,zc,3]T is an auxiliary variable vector. Moreover, to
simplify the tuning process, the observer gain vector can be
parameterized as
L ¼ ½2ωd;o ; ω2d;o T
where the observer bandwidth, ωd,o, is the only tunable parameter.
Finally, the control signal can also be designed as
α ¼ C P;PD  zd;3 =bd ;
where CP,PD is the original position PD controller.
Remark 2. In this situation, it can be thought that there is a
disturbance at the angular command αr, which brings about the
position steady-error. One of the functions of the observer is to
estimate this disturbance and then a corresponding opposite
control action can be executed.
4. Limit cycle analysis
Before incorporating the friction compensation approaches into
the control design, it is better to conduct limit cycle analysis to
verify whether the proposed control systems can generate limit
cycles or not. Here the limit cycle analysis is performed based on
the describing function method [25,26].
For the proposed reference compensation strategy, the closed-
loop control system consists of five parts as shown in Fig. 2: the
model from the angular output to the relative position of the ball,
the servo motor, the angular PD controller, the position PD
controller and the reference compensator, which are denoted as
Gm, Ga, CA,PD, CP,PD and CLADRC respectively.
Based on the assumptions and (1), the linear model of Gm could
be approximated as
Gm ðsÞ ¼ b0 =s2
The block diagram of Ga is shown in Fig. 3, where N(  ) is the
describing function of the Coulomb friction model, Kt is the motor
torque constant, Ke is the back electromotive force constant, J is the
rotor inertia, R and L are the resistance and inductance for the
equation at zero yields
ð22Þ  1=NðAÞ ¼ sðR þ LsÞ=ðK t C A;PD ð1 þ Gm C P;PD þGm C P;PD C PID Þ þ P L ðsÞÞ;
where
P L ðsÞ ¼ JLs3 þJRs2 þ K t K e s ð29Þ
Similarly, the equations related to the position PID control and the
direct compensation control schemes, both within the framework
ð23Þ
of the double loop cascade control structure as shown in Fig. 1, are
 1=NðAÞ ¼ sðR þ LsÞ=ðK t C A;PD ð1 þ Gm C P;PID Þ þ P L ðsÞÞ ð30Þ
and
ð24Þ
 1=NðAÞ ¼ sðR þ LsÞ=ðK t C A;PD ð1 þ Gm C P;Dir Þ þ P L ðsÞÞ ð31Þ
respectively, where CP,PID and CP,Dir represent the original position
PD controller adding an integral term and the direct compensator
respectively. For the Coulomb friction function, its describing
function is
 1=NðAÞ ¼  πA=ð4MÞ; ð32Þ
where M is the breakaway torque. The curve of  1/N(A) is just the
negative real axis. According to the describing function method,
the existence possibilities for limit cycles can be indicated by the
relative locations between the Nyquist curves for Eqs. (28), (30),
(31) and  1/N(A), respectively. For a typical friction model, the
breakaway torque M is the major characteristic parameter. Sub-
stituting s¼jω and letting ω vary from zero to 1, a locus for
common roots of the Nyquist curve can be plotted. The analysis
results will be presented in the following section.
5. Real paradigm with analysis, simulation and experiment
In this section, a ball and plate system, REBP450, is utilized as a
platform to validate the proposed method. The experimental setup
is shown in Fig. 4. The sampling frequency of the camera is 10 Hz.
Due to the low sampling rate, the current system can only achieve
ð25Þ
point positioning and it is unable to track a curve. The nominal
parameters of the plant are summarized in Tables 1 and 2 gives
the original angular and position PD control parameters.
5.1. Modeling of the original closed-loop position system

Before tuning the reference compensator, the model of the

Fig. 3. Block diagram of the servo motor.
original position closed-loop system should be obtained. As
described in subsection 3.1, this system can be represented by a
first-order model (5). Firstly, the nominal closed-loop step
response with a small amplitude command in the mathematical
simulation can be obtained. Then the parameters of model (5) can
be estimated using curve fitting [27]. The unit step responses are
shown in Fig. 5 when Km ¼ 1.11 and τ ¼0.58.
 Y. Wang et al. / ISA Transactions 53 (2014) 671–678 675

Nyquist Diagram
1.5

1 Direct compensation control

0.5

Imaginary Axis
0

−0.5
PID control

−1

−1.5
−1.4 −1.2 −1 −0.8 −0.6 −0.4 −0.2 0 0.2
Real Axis

Fig. 6. Nyquist plots for the linear part and  1/N(A): PID and direct compensation
control schemes.

Fig. 4. The ball and plate system: REBP450.

Km and τ into Eq. (8) yields b¼ 1.9. The bandwidth ωo of the
observer in the reference compensation should be determined
Table 1 carefully. Although a high ωo is helpful for achieving an accurate
Plant parameters.
estimation and a good compensation of disturbances, it may cause
Parameter Value high frequency oscillations in practice. Considering both the
robustness and the dynamic performance requirements, an appro-
m 0.05 kg priate tradeoff can be achieved by selecting ωo ¼ 1rad/s according
r 0.01 m
to the specified closed-loop bandwidth requirement [24]. The
g 9.81 m/s2
μ 0.0005 gains of the virtual control variable r0 are ke ¼3 and kv ¼ 0.4, tuned
Ib 2e-6 kg m2 by trial-and-error.
Ip 1.26 kg m2
Kt 0.03(Nm)A 5.3. Limit cycle analysis results
Ke 0.03(Nm)A
J 1.3 mg m2
R 4Ω In this subsection, the describing function method is used to
L 2.75 μH investigate whether the limit cycle will appear.
M 1 Firstly, a PID position controller, aiming at eliminating the
steady-error, is tested based on the original position PD control
parameters with an additional integral gain 0.12. Secondly, the
Table 2 direct compensation control scheme based on the disturbance
Control parameters.
observer is also investigated. The bandwidth of the observer, ωd,o,
Angular PD Controller Position PD Controller is selected as 2 rad/s. The Nyquist plots for the PID controller and
the direct compensation control as well as  1/N(A) are illustrated
proportion differential proportion differential as in Fig. 6. It reveals that an intersection point between the
Nyquist curve and  1/N(A) is ineluctable for either scheme, thus a
kA,p ¼ 65 kA,p ¼ 15 kP,p ¼4.5 kP,d ¼5
limit cycle is doomed to happening to some extent. In summary,
the double loop cascade control structure encounters a bottleneck.
For the proposed reference compensation approach, the
Nyquist plot and  1/N(A) are illustrated in Fig. 7. As the semi-
Nyquist curve does not intersect or contains  1/N(A), the limit
cycle could be avoided.

5.4. Simulation results

Before the control scheme is put into practice, it would be

Fig. 5. Responses of the original position control system and the identification model.
5.2. Tuning process for reference compensator
Based on the estimated model parameters, the reference
compensator can be tuned as follows. Substituting the values of
better to verify the proposed method using numerical simulations.
Firstly, the unit step responses of the system using the original
PD position control, the PID control and the direct compensation
control based on disturbance observer are shown in Fig. 8, respec-
tively. It should be noted that both the proportional and the
derivative gains are chosen as the same ones respectively as
provided by the developers to make the comparisons be fair. It is
visible that the steady-error cannot be eliminated with the original
PD control: the steady value of the real output is larger than the set-
point which is just due to the friction effect. When a compensation
mechanism is employed in the original position control location,
either using an integrator or using a disturbance observer, the
steady-error diminishes. However, there exists large overshoot and
long settling time with either scheme. These phenomena coincide
with the above mentioned limit cycle analysis perfectly. The contra-
diction between the steady-error elimination and the avoidance of
676 Y. Wang et al. / ISA Transactions 53 (2014) 671–678

Nyquist Diagram Direct compensation control

0.5
25 PID control
0.4
0.3
20
0.2
Imaginary Axis

0.1
15
0
−0.1

X(mm)
10
−0.2
−0.3 5
−0.4
−0.5 0
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2
Real Axis
-5
Fig. 7. Nyquist plots for the linear part and  1/N(A): reference compensation
control scheme. -10
0 10 20 30 40 50
Time(sec)

Fig. 10. Experimental results for 15 mm step response with PID and direct
compensation control in x-axis.

gain of the original closed-loop system is detected larger than 1,

a smaller command is produced to counteract this larger gain to
make the output just equal to the command. Therefore, a smooth
response can be obtained with neither steady-error nor overshoot,
which is difficult to obtain with the traditional double loop
cascade control structure.

5.5. Experimental results

In this sub-section, several comparative experiments were

Fig. 8. Simulation results for unit step response with PD, PID and direct compensa-
tion control schemes.
Fig. 9. Simulation results for unit step response with reference compensation
control scheme.
limit cycle is a severe dilemma for the traditional double loop
cascade control structure as shown in Fig. 1.
Next, the novel triple loop cascade control structure as shown
in Fig. 2 is investigated. The simulation results of the proposed
reference compensation control scheme are given in Fig. 9. As the
real output of the original system is larger than the command, it is
evident that a pseudo-reference command, which is r in Fig. 2 and
is smaller than the command, is generated to cancel the steady-
error resulted from the friction. In other words, when the steady
conducted to evaluate the performance of the reference compen-
sation methodology. It is especially difficult to achieve high-
precision control performance in the low-velocity regime due to
the friction effects. Therefore, a small position step command,
(15 mm, 15 mm), is chosen to validate the proposed method.
Firstly, the compensation methods based on the double loop
cascade control structure were tested. The experiments with the
PID control and the direct compensation control strategies are
shown in Fig. 10 and Fig. 11 for the two axes, respectively. The large
overshoot and long settling time are inevitable, matching both the
limit cycle analysis results shown in Fig. 6 and the mathematical
simulation results shown in Fig. 8. Moreover, the responses
possess visible flat segments in particular, the striking feature of
limit cycle in reality. Then experimental results of the reference
compensation control strategy proposed in this paper are shown
in Figs. 12,13 and for the two axes, respectively. With the help of
pseudo-reference compensation, the steady-error in the original
closed-loop position systems can be completely eliminated with-
out triggering the overshoots or the flat segments. It is also
obvious that the settling time is very short. As a result, the desired
high-precision and fast response objectives can be achieved by
using the proposed reference compensation strategy. Note that we
were unable to obtain the entirely identical initial conditions
because of the caliber errors under the dynamic circumstances
in the experiments and the low sampling frequency 10 Hz of the
position. The occasional tiny outliers in all the curves of the system
responses are just due to the positioning errors from the resolu-
tion of the digital camera.
6. Conclusion
In this paper, a reference compensation strategy, which is
independent of the friction model, has been proposed for controlling
a ball and plate system. A reduced-order observer was employed
to perform this reference compensation in a linear manner. The
 Y. Wang et al. / ISA Transactions 53 (2014) 671–678 677

Direct compensation control and direct compensation control strategies within the framework of
25 PID control the traditional double loop cascade control structure. Moreover, this
approach can also be applied to many other mechanical systems
20 operating in the low-velocity regime with high-precision require-
ments since it can reject the harmful effects caused by the friction.
15
Y(mm)

10
Acknowledgment
5
The authors thank the editors and the anonymous reviewers
0 for their helpful comments and suggestions that have improved
the presentation of this manuscript. This work was supported
-5
partly by the Natural Science Foundation of China under Grant
0 10 20 30 40 50 61174094 and 61273138, the Program for New Century Excellent
Time(sec) Talents in University under Grant no.NCET-10–0506, the Tianjin
Natural Science Foundation under Grant no. 13JCYBJC17400, the
Fig. 11. Experimental results for 15 mm step responses with PID and direct
China/South African Research Cooperation Program (No. 78673)
compensation control in y-axis.
and South Africa National Research Foundation Incentive (Grant
no. 81705).
25 Reference compensation control
PD control
Pesudo reference command
20 References

15 [1] Almutairi NB, Zribi M. On the sliding mode control of a ball on a beam system.
Nonlinear Dyn 2009;59(2):221–38 (June).
10 [2] Andrej Knuplez R.S., Chowdhury A. Modeling and control design for the ball
X(mm)

and plate system. In Proceedings of the IEEE International conference on

industrial technology, Maribor, Slovenia. vol. 2; December 2003. p. 1064–67.
5 [3] Gharieb W., Nagib G. Fuzzy intervention in PID controller design. In: Proceed-
ings of the IEEE International symposium on industrial electronics. vol. 3,
Pusan, Korea; 2001, p. 1639–43.
0
[4] Casagrande D, Astolfi A, Parisini T. Switching-driving Lyapunov function and
the stabilization of the ball-and-plate system. IEEE Trans Autom Control
-5 2009;54(8):1881–6 (August).
[5] Fan X, Zhang N, Teng S. Trajectory planning and tracking of ball and plate
system using hierarchical fuzzy control scheme. Fuzzy Sets Syst 2003;14
-10 (2):297–312 (June).
0 10 20 30 40 50 [6] Bai M., Lu H., Su J., Tian Y. Motion control of ball and plate system using
Time(sec) supervisory fuzzy controller. In: Proceedings of the 6th World congress on
intelligent control and automation.vol. 2, Dalian, China; June 2006. p. 8127–31.
Fig. 12. Experimental results for 15 mm step response with reference compensa- [7] Armstrong-Hélouvry B, Dupont P, Canudas de Wit C. A survey of models,
tion control in x-axis. analysis tools and compensation methods for the control of machine with
friction. Automatica 1994;30(7):1083–138 (July).
[8] Lin J, Chen CH. Positioning and tracking of a linear motion stage with friction
PD control compensation by fuzzy logic approach. ISA Trans 2007;46(3):327–42.
20 Reference compensation control [9] Yuan D., Jin H., Meng G., Liu C. Global sliding mode variable structure control
Pesudo reference command applied to pneumatic servo system. In: Proceedings of the IEEE International
18 symposium on knowledge acquisition and modeling; December 2008. p. 806–09.
[10] Friedland B, Park Y-J. On adaptive friction compensation. IEEE Trans Autom
16 Control 1992;37(10):1609–12.
14 [11] Bender FA, Lamaert V, Swevers J. The generalized Maxwell-slip model: a novel
model for friction simulation and compensation. IEEE Trans Autom Control
12 2005;50(11):1883–7.
[12] Olsson H, Äström KJ. Friction generated limit cycles. IEEE Trans Control Syst
10
Y(mm)

Technol 2001;9(4):629–36.
8 [13] Putra D, Nijmeijer H, van de Wouw N. Analysis of undercompensation and
overcompensation of friction in 1DOF mechanical system. Automatica 2007;43
6 (8):1387–94.
[14] Han J. From PID to active disturbance rejection control. IEEE Trans Ind Electron
4 2009;56(3):900–6.
[15] Sun D. Comments on active disturbance rejection control. IEEE Trans Ind
2
Electron 2007;54(6):3428–9.
0 [16] Goforth FJ, Zheng Q, Gao Z. A novel practical control approach for rate
independent hysteretic systems. ISA Trans 2012;51(3):477–84.
-2 [17] Dong L, Zhang Y, Gao Z. A robust decentralized load frequency controller for
0 10 20 30 40 50 interconnected power systems. ISA Trans 2012;51(3):410–9.
Time(sec) [18] Su YX, Duan BY, Zheng CH, Zhang YF, Chen GD, Mi JW. Disturbance rejection
high-precision motion control of a stewart platform. IEEE Trans Control Syst
Fig. 13. Experimental results for 15 mm step response with reference compensa- Technol 2004;12(3):364–74.
tion control in y-axis. [19] Sun B, Gao Z. A DSP-based active disturbance rejection control design for a 1-
kW H-bridge DC-DC power converter. IEEE Trans Ind Electron 2005;52
(5):1271–7.
[20] Li SL, Yang X, Yang D. Active disturbance rejection control for high pointing
describing function method validated that the limit cycle cannot accuracy and rotation speed. Automatica 2009;45(8):1854–60.
appear using the proposed reference compensation method. The [21] Zheng Q, Chen Z, Gao Z. A practical approach to disturbance decoupling
results of the mathematical simulations and practical experiments control. Control Eng Pract 2009;17(9):1016–25.
[22] Li S, Liu Z. Adaptive speed control for permanent-magnet synchronous motor
showed that the proposed method could achieve remarkable system with variations of load inertia. IEEE Trans Ind Electron 2009;56
performance improvement comparing with the original PD, PID (8):3050–9.
678 Y. Wang et al. / ISA Transactions 53 (2014) 671–678

[23] Li M, Li D, Wang J, et al. Active disturbance rejection control for fractional-
order system. ISA Trans 2013;52(3):365–74 (May).
[24] Gao Z. Scaling and bandwidth-parameterization based controller tuning. In:
Proceedings of American control conference. Denver, Colorado, USA; June
2003. p. 4989–96.
[25] Luo Y, Chen YQ, Pi Y. Fractional order ultra low-speed position servo:
Improved performance via describing function analysis. ISA Trans 2011;50
(1):53–60.
[26] Wang YJ. Robust prevention of limit cycle for nonlinear control systems with
parametric uncertainties both in the linear plant and nonlinearity. ISA Trans
2007;46(4):479–91.
[27] Bayard DS. High-order multivariable transfer function curve fitting: algo-
rithms, sparse matrix methods and experimental results. Automatica 1994;30
(9):1439–44.