Mechanical Systems and Signal Processing 109 (2018) 27–44

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Adaptive sliding mode control with moving surface:

Experimental validation for electropneumatic system
Assil Ayadi a,⇑, Mohamed Smaoui b, Sinda Aloui a, Soufien Hajji a, Mondher Farza c
a
Laboratory Lab-STA, LR11ES50, National School of Engineering of Sfax, University of Sfax, PB 1173, 3038 Sfax, Tunisia
b
Laboratoire Ampère, UMR CNRS 5005, INSA-Lyon, Université de Lyon, Lyon F-69621, Villeurbanne, France
c
Laboratoire GREYC, UMR CNRS 6072, ENSICAEN, Université de Caen, Caen 14032, France

a r t i c l e i n f o a b s t r a c t

Article history: In this work, an Adaptive Sliding Mode Control (ASMC) is proposed for a class of nonlinear
Received 26 August 2017 MIMO system with external disturbances. Although the Sliding Mode Control (SMC) is
Received in revised form 12 January 2018 known for its precision and robustness against disturbance and uncertainties, it suffers
Accepted 25 February 2018
from a chattering phenomenon and the choice of sliding surface parameters. To solve such
Available online 3 March 2018
problems, the SMC discontinuous term was replaced by a proportional derivative term and
a moving sliding surface was used. The adaptive parameters were obtained by Lyapunov
Keywords:
stability analysis to guarantee the stability of the closed loop system. In order to illustrate
Nonlinear MIMO system
Moving sliding surface
the efficiency of the proposed controller, experimental results on electropneumatic system
Adaptive sliding mode control were presented and compared to a classic sliding mode controller.
Electropneumatic system Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction

The sliding mode control is a nonlinear control technique featuring remarkable properties of accuracy, robustness and
easy tuning and implementation with a very large application fields [1,2]. Due to the use of discontinuous function and high
control gain, the main features of this kind of strategy are well presented with the robustness of closed-loop system and the
finite-time convergence. Concerning the field of application, many results have been published on sliding mode control
applied to pneumatic systems (see [3–6]). However, the major drawback of the sliding mode control is the chattering
phenomenon. Indeed, the discontinuous control term causes large oscillations around the sliding surface and leads to the
appearance of this phenomenon. The origin of this phenomenon is related to delays in control discontinuities since there
are no switching component elements capable of shifting to an infinite frequency [7].
In order to solve this problem, several techniques have been used. In [8–13], a High Order Sliding Mode Control (HOSMC)
has been proposed. This algorithm ensures that the sliding variable and its consecutive derivatives tend to zero in finite time
with the presence of uncertainties and disturbances. Since the HOSMC also allows increasing the sliding variable
stabilization accuracy, it is still frequently applied to the control of electropneumatic actuators.
Furthermore, many Adaptive Sliding Mode (ASMC) techniques have been used to reduce the chattering phenomenon
[14–20]. In fact, the controller conception does not need complete information about the uncertainty and perturbation
bounds due to the dynamic gains adaptation. These gains increase automatically resulting in dangerous oscillations because
of a too large switching control. Thus, the important feature of the adaptation algorithm is to ensure not to overestimate the

⇑ Corresponding author.
E-mail address: [email protected] (A. Ayadi).

https://doi.org/10.1016/j.ymssp.2018.02.042
0888-3270/Ó 2018 Elsevier Ltd. All rights reserved.
28 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

Nomenclature

y; v ; a position ðmÞ, velocity ðm=sÞ, acceleration ðm=s2 Þ

j jerk ðm=s3 Þ
pP;N pressure in the chamber P and N ðPaÞ
pE exhaust pressure ðPaÞ
uP ; uN servodistributors voltages ðVÞ
qm mass flow rate provided from servodistributor to cylinder chamber ðkg=sÞ
uð:Þ leakage polynomial function ðkg=sÞ
wð:Þ polynomial function ðkg=s=VÞ
bv viscous friction coefficient ðN=m=sÞ
F ext external force ðNÞ
k polytropic constant
l stroke length ðmÞ
M total load mass ðkgÞ
S piston P; N section
Ts temperature supply ðKÞ
r perfect gas constant related to unit mass ðJ=kg=KÞ

control gain values. In [17], the authors present an adaptive version of sliding mode control to track the position and pressure
of pneumatic actuator. Adaptation laws for the twisting and supertwisting algorithm have been reported in [14,15,21] to
track the position of electropneumatic system.
Moreover, [22,23] develop an ASMC for SISO system to track the position of the pneumatic actuator. The SMC switching
term is replaced by an adaptive proportional derivative term in order to attenuate the chattering phenomenon. However, the
validity of the control law is highly dependent on the stability of the unobservable one-dimensional subsystem. So, it is very
difficult to obtain results about the global stability of the zero dynamics. Then, the control of two different trajectories seems
possible if two servodistributors were used. For example, position and pressure could be controlled without any degradation
of the position tracking [4,24]. Besides, we have interest in choosing properly the coefficients of the sliding surface. In fact,
the sliding surface with coefficients to minimum values leads to solve errors convergence and longer tracking time. On the
other hand, sliding surface with coefficients to maximum values makes to faster errors convergence but the tracking occur-
rence can be degraded.
The objective of the current paper is to propose an adaptive sliding mode control with moving surface applied to an elec-
tropneumatic system in order to track the position and pressure. The main idea of this approach is to use, firstly, the adjus-
table coefficient of the sliding surface to solve the problem of reaching phase duration. In fact, the sliding surface is moved by
changing the magnitude of the slopes. Secondly, The adaptive PD term is introduced to reduce the chattering phenomenon.
All parameters, adaptive laws and sliding surface, are derived using Lyapunov stability analysis.
The remaining of this paper is organized as follows: Section 2 presents the description of nonlinear MIMO systems. In
Section 3, the classic sliding mode control is studied. An adaptive sliding mode controller with a moving sliding surface is
proposed and the stability analysis of the closed loop system is given in Section 4. Section 5 provides the description and
the modeling of the used pneumatic actuator system. In Section 6, the effectiveness of the proposed controller is checked
by an experimental comparative study with the classic sliding mode control. The conclusions on the developed work are
drawn and our potential future work is suggested in the final section of the paper.

2. Formulation

Consider a class of nonlinear MIMO systems described by the following dynamics equations:
8
> Xp
< xðri Þ ¼ f ðxÞ þ g ij ðxÞuj þ di ðtÞ
i i
ð1Þ
>
:
j¼1

yi ¼ xi ; i ¼ 1; . . . ; p
where f i ðxÞ and g ij ðxÞ are nonlinear functions assumed to be known and bounded 8i; j ¼ 1 . . . p;
h iT
ðr 1Þ T
x ¼ x1 ; x_ 1 ; . . . ; xðr
1
1 1Þ
; . . . ; xp ; x_ p ; . . . ; xp p 2 Rn is the state vector; u ¼ ½ u1 ; . . . ; up  2 Rp is the input vector;
 T
y ¼ y1 ; y2 ; . . . ; yp 2 Rp is the output vector; di ðtÞ is the unknown external disturbance assumed to have an upper bound Di .
Denote:
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 29

2 3
g 11 ðxÞ . . . g 1p ðxÞ
 T 6 . .. 7
FðxÞ ¼ f 1 ðxÞ; f 2 ðxÞ; . . . ; f p ðxÞ ; GðxÞ ¼ 6
4 .. . ... 75:
g p1 ðxÞ . . . g pp ðxÞ

h iT  T
ðrp Þ
xðnÞ ¼ xðr 1Þ ðr 2 Þ
1 ; x2 ; . . . ; xp
; D ¼ d1 ; d2 ; . . . ; dp :
P
We note that pi¼1 ni ¼ n with n is system dimension.
System (1) can be rewritten under the following compact form:

xðnÞ ¼ FðxÞ þ GðxÞu þ D ð2Þ

The synthesis of the control law, for the considered nonlinear systems class, requires the adoption of the following
assumptions:

Assumption 1. The matrix GðxÞ is invertible 8x and G1 ðxÞ is the inverse.

Assumption 2. The desired trajectories xdi ðtÞ; i ¼ 1 . . . p are functions of time known, bounded and have bounded
derivatives.
The tracking errors ei are given by the following relationship:
ei ¼ xi  xdi ; i ¼ 1; . . . ; p ð3Þ
Consider the sliding surfaces si , defined by:

X
ni
ðj1Þ
si ¼ cij ei ; i ¼ 1; . . . ; p ð4Þ
j¼1

with cini ¼ 1. To ensure the asymptotic stability, the coefficients ci1 ; . . . ; cini 1 must be positive constants which generate Hur-
witz polynomial.
In order to achieve the control objectives, we propose the vector of sliding surfaces S, defined by:

S ¼ ½ s1 s2 . . . sp T 2 R p ð5Þ

Moreover, the sliding condition is defined by:

1 d
ðsi Þ2 6 gi jsi j where gi > Di 8i ¼ 1 . . . p ð6Þ
2 dt
with gi are positive constants.
We assume that FðxÞ and GðxÞ are known, the SMC control law uwhich ensures the sliding condition is given by (6). This
control law consists of two terms: an equivalent control law ueq which ensures the sliding on the surface and a switching
control usw which brings the trajectories to the sliding surface and guarantees the robustness with respect to parametric vari-
ations and external disturbances [1,2].

u ¼ ueq þ usw ð7Þ

where ueq is the equivalent law which is determined by S_ ¼ 0 and without considering the effects of external disturbances.
" #
1 ðn Þ
X
nm
ueq ¼ G ðxÞ FðxÞ þ xd m  fj1 Eðj1Þ ð8Þ
j¼2

with nm ¼ max ni ; i ¼ 1; . . . ; p,
h i h iT
EðjÞ ¼ eðjÞ
p ðjÞ ðjÞ
fj ¼ diag c1j or ; c2j or ; . . . cj or ; ; 1 e2 . . . ep ; j ¼ 1; . . . ; nm  1;
h iT
ðn Þ ðn Þ ðn Þ ðn Þ
xd m ¼ xd11 xd22 . . . xdpp :

The switching law usw is chosen such that:

usw ¼ G1 ðxÞudis ð9Þ

where udis is given by:
30 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

2 3
g1 signðs1 Þ
6 7
6 g signðs Þ 7
6 2 2 7
6 7
udis ¼6
6
7
7 ð10Þ
6 .. 7
6 . 7
4 5
gp signðsp Þ
with gi are positive constants, i ¼ 1; . . . ; p.
Hence the control law given by (7) can be rewritten as follows:
" #
1 ðn Þ
X
nm
u ¼ G ðxÞ FðxÞ þ xd m  fj1 Eðj1Þ  udis ð11Þ
j¼2

The SMC is characterized by its precision and robustness to parametric variations and external disturbances. Yet, the main
disadvantage of this kind of control lies in the difficulty of choosing the sliding surface parameters, ie the robustness of the
system during the reaching phase and the chattering phenomenon.
In what follows, the ASMC where the surface is moving was introduced. This control technique is based on the choice of a
variant time linear sliding surface that conveniently adapts to arbitrary initial conditions and allows to obtain better tracking
performances. To attenuate the chattering phenomenon, an adaptive derivative proportional term was incorporated in the
global control law.

3. Adaptive sliding mode control

3.1. Adaptive sliding mode controller design

The discontinuous term is replaced by an adaptive PD term.

usw ¼ G1 ðxÞuPD ð12Þ

This adaptive term can be written as follows:
2 3
kp 1 s1 þ kd1 dtd s1
6 7
6 kp 2 s2 þ kd2 dtd s2 7
6 7
uPD ¼ qðS; hÞ ¼ 6
^ .. 7
6 . 7
4 5
kp p sp þ kdp dt sp
d

 
¼ q ^ 1 ðs1 ; h1 Þ; . . . ; q
^ p ðsp ; hp Þ T
h iT
¼ hT1 Wðs1 Þ; . . . ; hTp Wðsp Þ
¼ WðSÞh ð13Þ
 T h iT  T  
where hi ¼ kp i kdi ; h ¼ hT1 ; . . . ; hTp , Wðsi Þ ¼ si d
s
dt i
and WðSÞ ¼ diag WT ðs1 Þ; . . . ; WT ðsp Þ with kp i and kd i are the
adjusted gains, i ¼ 1; . . . ; p.
We define si the sliding moving surfaces defined by:
X
ni
ðj1Þ
si ¼ cimj ei i ¼ 1; . . . ; p ð14Þ
j¼1

with cm ini ¼ 1.
The vector of the moving sliding surfaces Sis defined by:
T
S ¼ ½ s1 ; s2 ; . . . ; sp  ð15Þ

The global control law is written as follows:

" #
X
nm
u ¼ G1 ðxÞ FðxÞ þ xd m 
ðn Þ ^fm Eðj1Þ  q
^ ðS; ^hÞ ð16Þ
j1
j¼2
h i
where ^fmj ¼ diag ^c1mj or ; ^c2m or ;
j
... ^cpmj or ; ; j ¼ 1 . . . nm  1.
The parameter vectors are computed using the following adaptation laws:
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 31

^f_ mi ¼ c SEðiÞT with i ¼ 1 . . . nm  1 ð17Þ

fi

^h_ ¼ c WT ðSÞS ð18Þ

h

with cfi > 0 and ch > 0 are the adaption gains.

The adaptive control schema (see Fig. 1) is described by the following block diagram:

3.2. Stability analysis

Define the optimal parameters vector:

!
hi ¼ arg min supkgi signðsi Þ  qi ðsi ; hi Þk ; i ¼ 1...p ð19Þ
hi 2Xi si 2R

where Xi are the constraints set for hi .

Assuming that the constraints set Xi are specified as follows:
Xi ¼ fhi : jjhi jj < M hi g ð20Þ
where M hi is a predefined parameter.
wi is the minimum of the approximation errors defined as:
wi ¼ gi signðsi Þ  qi ðsi ; hi Þ ð21Þ
Denote:
T
W PD ¼ ½ w1 ; . . . ; wp  ð22Þ
The approximation error vector belongs to L2 ½0; t; 8t P 0, with kW PD k 6 W PDmax .
Differentiating the moving sliding surface vector S with respect to time, given by (15), and using the control law u in (16),
the time derivative of the vector S can be described as follows:
X
nm X
nm
S_ ¼  ^fmðj1Þ Eðj1Þ  q
^ ðSj^hÞ þ fmðj1Þ Eðj1Þ þ D
j¼2 j¼2
nX
m 1

¼ ~fmj EðjÞ  WðSÞ^h þ d þ W PD  gsignðSÞ þ WðSÞh

j¼1
nX
m 1

¼ ~fmj EðjÞ þ WðSÞ~h þ W PD  gsignðSÞ þ D ð23Þ

j¼1

Fig. 1. Block diagram of adaptive sliding mode control with moving surface.
32 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

where ~fmj ¼ fmðj1Þ  ^fmðj1Þ and ~

h ¼ h  ^
h.
The Lyapunov function is chosen as follows:

nX
m 1
1 1 1
V¼ trðST SÞ þ trð~hT ~hÞ þ trð~fTmi~fmi Þ ð24Þ
2 2ch j¼1
2cfj
The time derivative of Vis given by:

nX
m 1
_ þ 1 trð~hT ~hÞ
V_ ¼ trðST SÞ
_
þ
1 _
trð~fTmi ~fmi Þ ð25Þ
ch j¼1
cfj
Replacing the time derivative of S given by (23) in (25), we get:
!!
nX
m 1

V_ ¼ tr ST ~fmj EðjÞ þ WðSÞ~h þ D þ W PD  gsignðSÞ

j¼1
nX
m 1
1 _ 1 _
þ trðh~T ~hÞ þ trð~fTmi ~fmi Þ ð26Þ
ch j¼1
cfj
nX
m 1

¼ trðST ~fj EðjÞ Þ þ trðST WðSÞ~hÞ  trðST gsignðSÞÞ

j¼1
nX
m 1
1 _ 1 _
þ trðST ðW PD þ DÞÞ þ trð~hT ~hÞ þ trð~fTmi~fmi Þ ð27Þ
ch j¼1
cfj
We note that:

nX
m 1 nX
m 1

trðST ~fj EðjÞ Þ þ trðST WðSÞ~hq Þ ¼ trðEðjÞT ~fTj SÞ þ trð~hTq WT ðSÞSÞ

j¼1 j¼1
X
nm
¼ trð~fTj SEðjÞT Þ þ trð~hTq WT ðSÞSÞ
j¼1

_ _ _ _ P
We have ~fj ¼ ^fj and ~
hq ¼ ^
hq and noting gs ¼ pi¼1 gi .
Hence, we can rewrite (27) as:
!!   
nX
m 1
1 _ 1 _
V_ ¼ tr ~fTmj SEðjÞT  ^fmj þ tr ~hT WT ðSÞS  ^h
j¼1
cfj ch
 gs trðS signðSÞÞ þ trðS ðW PD þ DÞÞ
T T
ð28Þ
By replacing (17) and (18) in (28), we obtain:

V_ ¼ gs ST signðSÞ þ ST ðD þ W PD Þ
V_ 6 gs jSj þ jSjðDsup þ W PDmax Þ ð29Þ

T Pp
with signðSÞ ¼ ½signðs1 Þ . . . signðsp Þ , jSj ¼ i¼1 jsi j and kDk 6 Dsup .
We can conclude that:

V_ 6 0; 8gs P W PDmax þ Dsup ð30Þ

Because V_ 6 0 is negative semi-definite i.e. VðtÞ 6 Vð0Þ implying that S; ~

h; ~f are bounded. From (23), it can be concluded
that S_ is bounded.
Integrating Eq. (29) from zero to t, it yields:
Z t
Vð0Þ P VðtÞ þ gs jSðsÞjds ð31Þ
0

Rt
Since Vð0Þ is bounded and VðtÞ is non increasing and bounded, it can be concluded that limt!1 0 jSðsÞjds is bounded. Then,
Rt
limt!1 0 jSðsÞjds is bounded and S_ is also bounded, based on Barbalat’s lemma, SðtÞ will converges asymptotically to zero
ðjÞ
and limt!1 SðtÞ ¼ 0. Then, ei converges to zero asymptotically.
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 33

Remark 1. The stability result is verified if all parameters involved in Eq. (25) are bounded. To ensure the boundedness of
this parameters, the adaptive laws (17) and (18) can be modified using the projection algorithm [25,26]. The modified
adaptive laws are given as follows.
For ^
hi , we use:
(
chi si Wðsi Þ if ðjj^hi jj < Mhi Þ or ðjj^hi jj ¼ M hi if si ^hTi Wðsi Þ P 0Þ
^h_ i ¼ ð32Þ
Phi ½chi si Wðsi Þ if ðjj^hi jj ¼ Mhi Þ and ðsi ^hTi Wðsi Þ > 0Þ

For ^cmj i , we use:

8
>
< cfi sj ej
ðiÞ ðiÞ
if ðjj^cmj i jj < M^c j Þ or ðjj^cmj i jj ¼ M^c j if sj ^cmj i ej P 0Þ
^c_ mj ¼
mi mi
ð33Þ
>
: Pc j ½cfi sj ejðiÞ  if ðjj^cmj i jj ¼ M^c j Þ and ðsj ^cmj i eðiÞ
j > 0Þ
i

mi mi

where M hi and M^c j are the design parameters that specify the allowable bounds of ^
hi and ^cmj i .
mi

The projection operator Phi ½ and Pc j ½ are defined as:
mi

^hi ^hT Wðsi Þ

Phi ½chi si Wðsi Þ ¼ chi si Wðsi Þ  chi si i
ð34Þ
jj^hi jj2
^cmj i ^cmj i eðiÞ
Pc j ½cfi sj ej  ¼ cfi sj ej  cfi sj
ðiÞ ðiÞ
ð35Þ
mi
jj^cmj i jj2

4. Application to pneumatic actuator position control

4.1. Pneumatic actuator modeling

The pneumatic actuator, considered in this study, is a linear double acting cylindric actuator which contains two cham-
bers denoted by P and N as shown in Fig. 2. Two three-way servo-distributors supply the air mass flow rates entering the two
chambers. The extremity of the cylinder axis carries a load of 17 kg on guiding rails. According to Newton’s second law
applied to the moving solid and under the assumptions of a link rod/ rigid carriage and a rigid body, the dynamic equation
of movement is presented by:
(
dy
¼v
dt
ð36Þ
dv
dt
¼ M1 ½SP pP  SN pN  F v is ðv Þ  F f ðv Þ  F ext 

where y and v are the position and the velocity of the rod, pP and pN are the pressures in chambers P and N, respectively.
F v is are viscous friction forces proportional to the velocity by a coefficient of viscosity bv :
F v is ¼ bv v ð37Þ

Fig. 2. Pneumatic actuator.

34 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

The dry friction forces F f , which act on the moving part in the presence of viscous friction, is a nonlinear model given by
several relations, for example [27]:
  
jv j
F f ¼ F C þ ðF S  F C Þexp  signðv Þ ð38Þ
c
where F S ; F C and c are the stiction, the Coulomb friction and the Stribeck constant effect.
The external forces F ext are considered constant. They represent the action of the atmospheric pressure PE on the moving
part.
F ext ¼ ðSP  SN ÞPE ð39Þ
To obtain the pressure evolution with variable volume, we assume that:

 the air is supposed to be a perfect gas,

 the pressure and the temperature are homogeneous in each chamber,
 the process is polytropic and it is characterized by a constant coefficient k,
 the temperature in both chambers is equal to the supplied one since its variation of temperature is negligible.

Since the servodistributors dynamic can be neglected, their model can be reduced to a static one and can be described by
two relationships qmP ðuP ; pP Þ and qmN ðuN ; pN Þ. The pressure dynamics in both chambers P and N are as follows:
8 h i
< dt ¼ V P ðyÞ qmP ðuP ; pP Þ  rT s pP v
>
>
dpP krT s SP

h i ð40Þ
>
>
: dpN ¼ krT s q ðuN ; p Þ þ SN p v
dt V N ðyÞ mN N rT s N

V P and V N are, respectively, the volumes of the P and N chambers:

(
V P ðyÞ ¼ V DP þ SP 2l þ SP y
V N ðyÞ ¼ V DN þ SN 2l  SN y

where V DP and V DN represent the actuator dead volumes.

4.2. Control model

In [28], the authors have proposed analytical models for both simulation and control purposes. The flow stage character-
istics were approximated by an affine polynomial functions in control such that:
qm ðu; pÞ ¼ uðpÞ þ wðp; signðuÞÞu ð41Þ
The polynomial functions uðpÞ; wðp; u > 0Þ and wðp; u < 0Þ degrees are equal to five. The maximum error of mass flow rate
between the polynomial approximation and the static measurement are less then 10%.

Assumption 3. F S and F C are not easily measurable and variable during experimental tests. Thus, only for the control model,
the dry friction forces F f ðv Þ in Eq. (38), have been neglected compared to other forces.

The nonlinear affine model (42) is obtained by referring to Eq. (36), (40) and (41) and taking account Assumption 3.
8 dy
>
> ¼v
> dt
>
>
>
>
>
>
>
>
dv
¼ 1 ½S p  SN pN  bv v  F ext 
< dt M P P
h i ð42Þ
>
> dpP
¼ VkrT uðpP Þ  rTSPs pP v þ VkrT wðpP ; signðuP ÞÞuP
>
>
s
P ðyÞ
s
P ðyÞ
>
>
dt
>
>
>
> h i
>
: dpN ¼ krT s uðp Þ þ SN p v þ krT s wðp ; signðuN ÞÞuN
dt V N ðyÞ N rT s N V N ðyÞ N

where k is the polytropic coefficient.

Including two inputs up and uN , the nonlinear model of the electropneumatic system has the following form:
x_ ¼ f ðxÞ þ gðxÞU ð43Þ

with x ¼ ½ y v pP T
pN  is the state vector, U ¼ ½ uP T
uN  is the input control and f ðxÞ and gðxÞ are the nonlinear functions.
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 35

where
0 1
v
B1 C
B M ½SP pP  SN pN  bv v  F ext  C
B h i C
B
f ðxÞ ¼ B C
B V
krT s
P ðyÞ
u ðpP Þ  SP
rT P
pP v C
C
@ h i A
krT s
V N ðyÞ
uðp N Þ þ SN
p
rT N N
v

gðxÞ ¼ ð g 1 ðxÞ g 2 ðxÞ Þ

0 1
0 0
B 0 0 C
B C
¼BB VkrTðyÞs wðpP ; sgnðuP ÞÞ 0
C
C
@ P A
0 krT s
V N ðyÞ
wðpN ; sgnðuN ÞÞ

The use of two servodistributors generates a system with two degrees of freedom via the control. This characteristic was
exploited so that the two control objectives are obtained. The supplementary degree of freedom issued from this new design
leads to control another output.
Let’s define hðxÞ the vector constituted of the two chosen outputs: position and pressure in chamber P:
   
h1 ðxÞ y
hðxÞ ¼ ¼ ð44Þ
h2 ðxÞ pP
In order to use a sliding mode technique, a coordinate transformation has been proposed in [4] with a diffeomorphism
given by (45). The nonlinear affine model is then given by (46).
0 1
h1 ðxÞ ¼ y
B Lf h1 ðxÞ ¼ v C
B C
z ¼ /ðxÞ ¼ B 2 C ð45Þ
@ Lf h1 ðxÞ ¼ a A
h2 ðxÞ ¼ pP
8 dy
>
> ¼v
> dt
>
>
> dv
> dt ¼ a
>
<
da
¼ L3f h1 ð/1 ðzÞÞ þ Lg1 L2f h1 ð/1 ðzÞÞup ð46Þ
>
>
dt
>
> þLg2 L2f h1 ð/1 ðzÞÞuN
>
>
>
>
: dpp
dt
¼ Lf h2 ð/1 ðzÞÞ þ Lg1 h2 ð/1 ðzÞÞuP
where

L3f h1 ðxÞ ¼ krT
M
s Sp
V p ðyÞ
uðpP Þ  V NSNðyÞ uðpN Þ
 Mbv2 ðSP pP  SN pN  bv v  F ext Þ ð47Þ
 2
S2 pN
 kMv V Pp ðyÞP þ VNN ðyÞ
S p

krT s Sp

Lg1 L2f h1 ðxÞ ¼ w pP ; sgnðup Þ > 0 ð48Þ
MV p ðyÞ
krT s SN
Lg2 L2f h1 ðxÞ ¼  wðpN ; sgnðuN ÞÞ < 0 ð49Þ
MV N ðyÞ
krT s kS
Lf h2 ðxÞ ¼ uðpP Þ  P pP v ð50Þ
V P ðyÞ V P ðyÞ
krT s

Lg1 h2 ðxÞ ¼ w pP ; sgnðup Þ > 0 ð51Þ
V p ðyÞ

Remark 1. The main control objective is to have a good accuracy in terms of position and pressure tracking. The position
relative degree is three and that of the pressure is one. Taking into account the previous information, the position and
pressure desired trajectories are at least three times and one time differentiable, respectively.
36 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

4.3. SMC

Two sliding surfaces s1 and s2 were defined by the following expressions:

s1 ¼ c1 e1 þ c2 e_ 1 þ €e1 ð52Þ
s2 ¼ e2 ð53Þ
where
e1 ¼ y  yd
e_ 1 ¼ y_  y_ d ¼ v  v d
€e1 ¼ y €y €d ¼ a  ad
e2 ¼ pP  pP d
yd ; v d ; ad and pP d are respectively the position, velocity, acceleration and pressure desired trajectories.
The sliding surface vector S is defined by:
 
s1
S¼ ð54Þ
s2
The sliding mode control law is given by the following equation:
" #
1 ðn Þ
X
nm
u ¼ G ðxÞ FðxÞ  hd m þ fj1 Eðj1Þ  udis ð55Þ
j¼2
     
L3f h1 ðxÞ L L2 h ðxÞ Lg2 L2f h1 ðxÞ uP
where FðxÞ ¼ , GðxÞ ¼ g1 f 1 ,u¼
Lf h2 ðxÞ Lg1 h2 ðxÞ 0 uN
" #T
c1j 0 h iT h iT
ðn Þ
fj ¼ diag EðjÞ ¼ eðjÞ
1 0 ; hd m ¼ yð3Þ
d p_ Pd ; nm ¼ 3
0 0

where udis is given by:

 
g1 signðs1 Þ
udis ¼ ð56Þ
g2 signðs2 Þ
g1 and g2 are positive constants.

Remark 2. It is worth noting that Lg 1 h2 is strictly positive and Lg2 L2f h1 is strictly negative, therefore there is no singularity in
the matrix GðxÞ.

4.4. ASMC

The moving sliding surfaces are defined by the following expressions:

s1 ¼ c1m1 e1 þ c1m2 e_ 1 þ €e1 ð57Þ

s2 ¼ e2 ð58Þ

where c1m1 et c1m2 are the sliding surface coefficients calculated from an adaptive algorithm to obtain a good tracking
performance.
We define the adaptive term uPD which is substituted to the discontinuous term in order to reduce chattering.
uPD ¼ WðSðtÞÞh ð59Þ
T
where h and WðSðtÞÞ are, respectively, the adjustable parameters vector and the regressive vector, defined by h ¼ ½hT1 hT2  and
T
WðSÞ ¼ diag½WT ðs1 Þ WT ðs2 Þ where hi ¼ ½kpi kdi T and Wðsi Þ ¼ ½si ðtÞ d
s ðtÞ ;
dt i
8i ¼ f1; 2g.
Therefore, the resulting ASMC law is expressed by:
" #
X
3
1
u ¼ G ðxÞ FðxÞ þ
ðn Þ
hd m  ^
fmj1 Eðj1Þ ^
 WðSðtÞÞh ð60Þ
j¼2

where ^fm1 ; ^fm2 and ^

h are the estimates of fm1 ; fm2 and h, respectively, defined as follows:
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 37

^f_ m1 ¼ c SE_ T ð61Þ

f1

^f_ m2 ¼ c SE€T ð62Þ

f2

^h_ ¼ c WT ðSÞS ð63Þ

h

where cf1 ; cf2 and ck are positive constants.

5. Experimental results

The experimental setup used to validate the proposed method is shown in Fig. 3. The system consists of two three way
proportional servodistributors (Servotronic Joucomatic); linear double acting cylindric actuator; controller board (dSpace
DS1104); position sensor (NovoTECHNIK, model TLH500); pressure sensors(KULITE model XT140M-7BarA) and PC.
The actuator under consideration is an inline electropneumatic cylinder using a simple rod (32=20 mm) with a stroke of
500 mm. The rod is connected to one side of a carriage and drives an inertial load on guiding rails. The total moving mass is
17 kg.
The adaptive moving sliding mode controller applied to an electropneumatic system has been implemented using dSpace
DS1104 controller board with a dedicated digital signal processor. The measured signals are all analogical. They are trans-
mitted via a signal conditioning unit before being converted by a 16-bits Analog Digital Converter ADC. We chose a sampling
time equal to 1 ms. The experimental results were discussed to demonstrate the effectiveness of the adaptive controller.
There are two ways to get some velocity feedback: using a velocity sensor or recovering the velocity signal using position
information. To have the minimum system cost, the cylinder velocity is obtained firstly by differentiating the position ana-
logically and then using a low-pass filter. All control laws are implemented using a position sensor and two pressure sensors.
Theirs characteristics are given in Table 1.
Table 2 shows the specifications of the pneumatic actuator.
The experiment results related to the pneumatic system controllers are given in Figs. 4–17. In order to show the proposed
controller efficiency, the experiment results of the SMC (Figs. 4–9) were compared with those of the ASMC (Figs. 10–17).
The implementation of the SMC law depends, necessarily, on the choice of the sliding surface coefficients c1 and c2 and
the switching gain values g1 and g2 . The coefficients of the sliding surface are chosen such that they generate a polynomial
Hurwitz and ensure a good tracking of the position. The coefficients c1 and c2 are fixed to 400 and 40, respectively. For the
switching gains, we chose g1 ¼ 70 and g2 = 20,000.

Fig. 3. Experimental setup.

38 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

Table 1
Sensors characteristics.

Sensor Number Mark Model Precision

Position 1 NovoTECHNIK TLH500 0:05%
Pressure 2 KULITE XT140M 700 Pa (0.1%)

Table 2
Specifications of the pneumatic actuator.

Parameter Value.Unit
d 3:2102 m
SP 7:2727104 m2
SN 4:909104 m2
V DP 1105 m2
V DN 1:2105 m2
l 0:5 m

Fig. 4. Position and desired position (m) (SMC).

Fig. 5. Position error (mm) (SMC).

 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 39

Fig. 6. Pressure and desired pressure (bar) (SMC).

Fig. 7. Pressure error (bar) (SMC).

From the experimental results, the SMC law presents a good performance for tracking the actuator position and regulat-
ing the chamber P pressure. In fact, Figs. 4 and 5 show the evolution of the controlled position, desired position and position
error. The maximum dynamic error is about 2:6 mm (1% of the total displacement magnitude). Furthermore, as shown in
Fig. 7, the maximum pressure error is 100 mbar and the mean absolute error is about 10 mbar. However, this control law
suffers from a major drawback which is the chattering phenomenon. The control inputs uP and uN given, respectively in
Figs. 8 and 9, present high oscillations. The presence of the chattering phenomenon is due to the presence of the sign function
in the control law. This phenomenon is undesirable for an electropneumatic system and can be responsible the short life
span of some components: especially the servodistributor.
Based on the experimental results, the ASMC has improved the tracking performance of the electropneumatic system and
reduced the chattering phenomenon. As shown in Figs. 10 and 11, we obtained a better position tracking than the SMC. The
maximal error does not exceed 1:35 mm ie 0:54% of the total displacement magnitude. Further, the proposed controller
guarantees a good regulation of pressure. In Fig. 13, the maximum pressure error is about 110 mbar which is the same using
the SMC. Using the same benchmark and in the same practical condition and setup, different control strategies have been
implemented to track the position or both the position and pressure. In [29], a backstepping control was developed to track
the position and the pressure. The maximum dynamic error is about 1:62 mm which is about 0:65% of the total displacement
and in the steady state, the position error is about 100 lm. The dynamic pressure error is about 5 mbar. Furthermore, a high
order sliding mode control was proposed to reduce the chattering and to track the position [9]. The maximum error in
dynamic stage is about 1:9 mm and it is about 0:11 mm. Moreover, in Figs. 14 and 15, we present the evolution of the control
40 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

Fig. 8. Input control uP (V) (SMC).

Fig. 9. Input control uN (V) (SMC).

Fig. 10. Position and desired position (m) (ASMC).

A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 41

Fig. 11. Position error (mm) (ASMC).

Fig. 12. Pressure and desired pressure (bar) (ASMC).

Fig. 13. Pressure error (bar) (ASMC).

42 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

Fig. 14. Input control uP (V) (ASMC).

Fig. 15. Input control uN (V) (ASMC).

input uP and uN , respectively. The reduction of the chattering phenomenon can be easily noticed especially at the static stage.
Figs. 16 and 17 illustrate the evolution of the adaptation parameters and sliding surface gains. From this point of view, the
proposed controller solves the problem caused by the SMC and seems more useful. Table 3 presents the maximum dynamic
error (MDE) and mean absolute error (MAE) of position and pressure using the SMC and the ASMC.

6. Conclusion

In this study, an adaptive sliding mode control was proposed for a class of nonlinear MIMO system with external distur-
bances. A moving surface was introduced and the SMC switching term is replaced by an adaptive PD term to obtain the best
sliding surface parameters and reduce the chattering phenomenon. All the sliding surface and PD term parameters are cal-
culated on line using an adaptive algorithm based on Lyapunov synthesis. The closed loop system stability is guaranteed by a
Lyapunov sense to ensure the convergence of the tracking error to zero. Finally, the proposed controller is applied to an elec-
tropneumatic system to track the position and pressure. The experimental results show that the adaptive sliding mode con-
trol reduces the chattering phenomenon and improves the tracking performances compared to the classic sliding mode
control.
 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44 43

Fig. 16. Control parameters gains (ASMC).

Fig. 17. Surface gains (ASMC).

Table 3
MDE and MAE error for SMC and ASMC.

Control law Position error (mm) Pressure error (bar)

MDE MAE MDE MAE
SMC 2.6 0.342 0.1 0.01
ASMC 1.35 0.235 0.11 0.015

Acknowledgement

This work was supported by the Ministry of Higher Education and Scientific Research of Tunisia.

References

[1] J.J. Slotine, Applied Nonlinear Control, Prentice-Hall. Inc., 1991.

[2] H.K. Khalil, Nonlinear Systems, Prentice Hall, 1996.
[3] M. Bouri, D. Thomasset, Sliding control of an electropneumatic actuator using an integral switching surface, IEEE Trans. Control Syst. Technol. 9 (2)
(2001) 368–375.
[4] M. Smaoui, X. Brun, D. Thomasset, Systematic control of an electropneumatic system: Integrator backstepping and sliding mode control, IEEE Trans.
Control Syst. Technol. 14 (5) (2006) 905–913.
44 A. Ayadi et al. / Mechanical Systems and Signal Processing 109 (2018) 27–44

[5] J.F. Carneiro, F.G. de Almeida, Accurate motion control of a servopneumatic system using integral sliding mode control, Int. J. Adv. Manuf. Technol. 77
(9–12) (2015) 1533–1548.
[6] M. Rahmani, A. Ghanbari, M.M. Ettefagh, Hybrid neural network fraction integral terminal sliding mode control of an inchworm robot manipulator,
Mech. Syst. Signal Process. 80 (2016) 117–136.
[7] K. Young, V. Utkin, U. Ozguner, A control engineer’s guide to sliding mode control, IEEE Trans. Control Syst. Technol. 7 (3) (1999) 328–342.
[8] S. Laghrouche, M. Smaoui, F. Plestan, X. Brun, Higher order sliding mode control based on optimal approach of an electropneumatic actuator, Int. J.
Control 79 (2) (2006) 119–131.
[9] M. Smaoui, X. Brun, D. Thomasset, High-order sliding mode for an electropneumatic system: a robust differentiator-controller design, Int. J. Robust
Nonlinear Control 18 (4–5) (2008) 481–501.
[10] A. Girin, F. Plestan, A. Glumineau, X. Brun, High order sliding mode controllers of an electropneumatic actuator: application to an aeronautic
benchmark, IEEE Trans. Control Syst. Technol. 17 (3) (2009) 633–645.
[11] F. Plestan, E. Moulay, A. Glumineau, Output feedback sampling control: a robust solution based on second order sliding mode, in: Decision and Control,
2009 held jointly with the 2009 28th Chinese Control Conference. CDC/CCC 2009. Proceedings of the 48th IEEE Conference on, 2009, pp. 8423–8427.
[12] X. Yan, M. Primot, F. Plestan, Electropneumatic actuator position control using second order sliding mode, e & i Elektrotech. Inform. 133 (6) (2016)
266–273.
[13] Z. Su, H. Wang, N. Li, Y. Yu, J. Wu, Exact docking flight controller for autonomous aerial refueling with back-stepping based high order sliding mode,
Mech. Syst. Signal Process. 101 (2018) 338–360.
[14] M. Taleb, A. Levant, F. Plestan, Pneumatic actuator control: solution based on adaptive twisting and experimentation, Control Eng. Pract. 21 (5) (2013)
727–736.
[15] Y. Shtessel, M. Taleb, F. Plestan, A novel adaptive-gain supertwisting sliding mode controller: methodology and application, Automatica 48 (5) (2012)
759–769.
[16] J. Fei, H. Ding, Adaptive sliding mode control of dynamic system using rbf neural network, Nonlinear Dyn. 70 (2) (2012) 1563–1573.
[17] F. Plestan, Y. Shtessel, V. Brageault, A. Poznyak, Sliding mode control with gain adaptation: application to an electropneumatic actuator, Control Eng.
Pract. 21 (5) (2013) 679–688.
[18] J. Fei, C. Lu, Adaptive sliding mode control of dynamic systems using double loop recurrent neural network structure, IEEE Trans. Neural Networks
Learn. Syst. PP (99) (2017) 1–12.
[19] S.-Y. Chen, S.-S. Gong, Speed tracking control of pneumatic motor servo systems using observation-based adaptive dynamic sliding-mode control,
Mech. Syst. Signal Process. 94 (2017) 111–128.
[20] S. Kang, H. Yan, L. Dong, C. Li, Finite-time adaptive sliding mode force control for electro-hydraulic load simulator based on improved gms friction
model, Mech. Syst. Signal Process. 102 (2018) 117–138.
[21] X. Yan, A. Estrada, F. Plestan, Adaptive pulse output feedback controller based on second-order sliding mode: methodology and application, IEEE Trans.
Control Syst. Technol. 24 (6) (2016) 2233–2240.
[22] A. Ayadi, S. Hajji, M. Smaoui, A. Chaari, Chattering-free adaptive sliding mode control for pneumatic system position tracking, in: 2015 16th
International Conference on Sciences and Techniques of Automatic Control and Computer Engineering (STA), 2015, pp. 752–757.
[23] A. Ayadi, S. Hajji, M. Smaoui, A. Chaari, Adaptive moving sliding mode control for siso systems: application to an electropneumatic system, J. Control
Sci. Eng. 2016 (2016) 10.
[24] A. Ayadi, S. Hajji, M. Smaoui, A. Chaari, M. Farza, Experimental sensorless control for electropneumatic system based on high gain observer and
adaptive sliding mode control, Int. J. Adv. Manuf. Technol. 93 (9) (2017) 4075–4088.
[25] H. Ho, Y. Wong, A. Rad, Adaptive fuzzy sliding mode control with chattering elimination for nonlinear {SISO} systems, Simul. Model. Pract. Theory 17
(7) (2009) 1199–1210.
[26] I.F. Jasim, P.W. Plapper, H. Voos, Adaptive sliding mode fuzzy control for unknown robots with arbitrarily-switched constraints, Mechatronics 30
(2015) 174–186.
[27] A. Tustin, The effects of backlash and of speed-dependent friction on the stability of closed-cycle control systems, J. Inst. Electr. Eng.-Part IIA: Automat.
Regul. Servo Mech. 94 (1) (1947) 143–151.
[28] M. Belgharbi, D. Thomasset, S. Scavarda, S. Sesmat, Analytical model of the flow stage of a pneumatic servodistributor for simulation and nonlinear
control, in: Sixth Scandinavian International Conference on Fluid Power, Tampere, Finland, 1996, pp. 847–860.
[29] M. Smaoui, X. Brun, D. Thomasset, A study on tracking position control of an electropneumatic system using backstepping design, Control Eng. Pract.
14 (8) (2006) 923–933.