Automatica 48 (2012) 759769

Contents lists available at SciVerse ScienceDirect

Automatica
journal homepage: www.elsevier.com/locate/automatica

A novel adaptive-gain supertwisting sliding mode controller: Methodology

and application
Yuri Shtessel a , Mohammed Taleb b,c , Franck Plestan b
a

The University of Alabama in Huntsville, Huntsville, USA

LUNAM Universit, Ecole Centrale de Nantes, IRCCyN UMR CNRS 6597, Nantes, France

Moulay Ismail University, Meknes, Morocco

article

info

Article history:
Received 20 January 2011
Received in revised form
12 July 2011
Accepted 28 September 2011
Available online 20 March 2012
Keywords:
Second order sliding mode
Adaptive control
Electropneumatic actuator

abstract
A novel super-twisting adaptive sliding mode control law is proposed for the control of an
electropneumatic actuator. The key-point of the paper is to consider that the bounds of uncertainties and
perturbations are not known. Then, the proposed control approach consists in using dynamically adapted
control gains that ensure the establishment, in a finite time, of a real second order sliding mode. The
important feature of the adaptation algorithm is in non-overestimating the values of the control gains.
A formal proof of the finite time convergence of the closed-loop system is derived using the Lyapunov
function technique. The efficiency of the controller is evaluated on an experimental set-up.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
During the past two decades, the development of highperformance linear/nonlinear controllers (Brun, Sesmat, Thomasset, & Scavarda, 1999; Brun, Thomasset, & Bideaux, 2002; Chiang,
Chen, & Tsou, 2005; Edge, 1997; Hamiti, Voda-Besanon, & RouxBuisson, 1996; Kimura, Hara, Fujita, & Kagawa, 1997; Kyoungkwan & Shinichi, 2005; Ming-Chang & Shy-I, 1995; Miyajima, Fujita,
Sakaki, Kawashima, & Kagawa, 2007; Rao & Bone, 2006; Richard &
Scavarda, 1996; Schultea & Hahn, 2004; Smaoui, Brun, & Thomasset, 2006) yields the possibility of reaching high accuracy positioning for a pneumatic actuator. However, due to uncertainties,
robust controllers are necessary to ensure positioning with high
precision. In this respect, sliding mode controllers are used for electropneumatic actuators (Bouri & Thomasset, 2001; Paul, Mishra,
& Radke, 1994; Smaoui, Brun, & Thomasset, 2005; Yang & Lilly,
2003). Sliding mode control is one of the best choices for controlling perturbed systems with matched disturbances/uncertainties
(Edwards & Spurgeon, 1998; Utkin, Guldner, & Shi, 1999). The price

This work has been partially supported by CNRS through Musclair PEPS
project. The material in this paper was partially presented at the 18th IFAC
World Congress, August 28September 2, 2011, Milano, Italy. This paper was
recommended for publication in revised form by Associate Editor Warren E. Dixon
under the direction of Editor Andrew R. Teel.
E-mail addresses: shtessel@ece.uah.edu (Y. Shtessel),
Mohammed.Taleb@irccyn.ec-nantes.fr (M. Taleb),
Franck.Plestan@irccyn.ec-nantes.fr (F. Plestan).

0005-1098/$ see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.automatica.2012.02.024

for achieving the robustness/insensitivity to these disturbances is

control chattering (Boiko, 2008; Edwards & Spurgeon, 1998; Fridman, 2001, 2002; Utkin et al., 1999). The traditional ways for reducing chattering are as follows:
(a) Replacing the discontinuous control function by saturation
or sigmoid ones (Burton & Zinober, 1986; Slotine & Li,
1991). This approach yields continuous control and chattering
elimination. However, it constrains the sliding systems
trajectories not to the sliding surface but to its vicinity losing
the robustness to the disturbances.
(b) Using the higher order sliding mode control techniques
(Djema, Barbot, & Busawon, 2008; Laghrouche, Plestan, & Glumineau, 2007; Laghrouche, Smaoui, Plestan, & Brun, 2006;
Levant, 2003, 2005; Plestan, Glumineau, & Laghrouche, 2008;
Shtessel, Shkolnikov, & Levant, 2007). This approach allows
driving to zero the sliding variable and its consecutive derivatives in the presence of the disturbances/uncertainties increasing the accuracy of the sliding variable stabilization, and has
still been successfully applied for the control of electropneumatic actuators (Girin & Plestan, 2009; Laghrouche et al., 2006).
However, the main challenge of high order sliding mode controllers is the use of high order time derivatives of the sliding variable. It is worth noting that some second order sliding
mode control, the popular super-twisting algorithm (Levant,
1993) and gain-commuted controller (Plestan, Moulay, & Glumineau, 2010a) only require measurement of the sliding variable whereas the other second order sliding mode controllers
also need the time derivative of the sliding variable.

760

Y. Shtessel et al. / Automatica 48 (2012) 759769

(c) Using controllers with dynamical gains. Recently, adaptive

sliding mode controllers have been proposed, the interest
being the adaptation of the gain magnitude with respect to
uncertainty/perturbation effects. Then, a reduced gain induces
lower chattering. In Plestan, Shtessel, Brgeault, and Poznyak
(2010b), an adaptive (first order) sliding mode controller has
been proposed and has been evaluated for the control of an
electropneumatic actuator.
The main objective of this paper is to bring together two of the
previous chattering reduction approaches, gain adaptation and
high order sliding mode control. The obtained controller, based
on the well-known super-twisting (Levant, 1993) second order
sliding mode algorithm, does not require any information on the
boundaries of the disturbance and its gradient except for their
existence. It will yield to the very first application to a real system
of such method.
The super-twisting control law (STW) is one of the most
powerful second order continuous sliding mode control algorithms
that handles a relative degree equal to one. It generates the
continuous control function that drives the sliding variable and
its derivative to zero in finite time in the presence of the
smooth matched disturbances with bounded gradient, when this
boundary is known. Since STW algorithm contains a discontinuous
function under the integral, chattering is not eliminated but
attenuated. The main disadvantage of STW control algorithm
is that it requires the knowledge of the boundaries of the
disturbance gradient. In many practical cases this boundary
cannot be easily estimated. The overestimating of the disturbance
boundary yields to larger than necessary control gains, while
designing the STW control law. The adaptive-gain STW (ASTW)
control law, which handles the perturbed plant dynamics with
the additive disturbance/uncertainty of certain class with the
unknown boundary, was proposed in Shtessel, Moreno, Plestan,
Fridman, and Poznyak (2010). In this latter paper, a novel adaptive
STW control law that continuously drives the sliding variable and
its derivative to zero in the presence of the bounded disturbance
with the unknown boundary, has been proposed. The finite
convergence time is estimated. The proof is based on recently
proposed Lyapunov function (Moreno & Osorio, 2008; Polyakov
& Poznyak, 2009). The current paper is extending the result of
Shtessel et al. (2010) to a larger class of nonlinear uncertain
systems.
This paper is organized as follows. Section 2 gives the
main methodological results concerning the novel adaptive
super-twisting algorithm. Section 3 describes the experimental
set-up composed by two electropneumatic actuators. Then,
Section 4 described the position controller of one of the both
electropneumatic actuators (the second being controlled in force
by an another way), and displays obtained experimental results.
2. Novel adaptive supertwisting controller design
2.1. Problem statement

Therefore, the inputoutput dynamics can be presented

+
f (x) +
g (x)u
t x x
a(x,t )

b(x,t )

= a(x, t ) + b(x, t )u.

(2)

Also, it is assumed that

A3. The function b(x, t ) R is uncertain and can be presented

as
b(x, t ) = b0 (x, t ) + b(x, t )

(3)

where b0 (x, t ) > 0 is a known function and b(x, t ) is a

bounded perturbation so that

|b(x, t )|
= (x, t ) 1 < 1
b0 ( x, t )
x Rn and t [0, ) with an unknown boundary 1 .
A4. The function a(x, t ) R is presented as
a(x, t ) = a1 (x, t ) + a2 (x, t )

(4)

with the bounded terms

|a1 (x, t )| 1 | |1/2

|a2 (x, t )| 2

(5)

where the finite boundaries 1 , 2 > 0 exist but are not known.
Finally, one gets

b(x, t )

= a(x, t ) + 1 +
b0 ( x, t )

(6)

b1 (x,t )

where = b0 (x, t )u. From A3, one gets

A5.
1 1 b1 (x, t ) 1 + 1 .

(7)

The problem is to drive the sliding variable and its derivative

to zero in finite time in the presence of the bounded additive (5)
and multiplicative (3) perturbations with the unknown boundaries
1 , 2 , 1 > 0 by means of continuous control without the control
gain overestimation.
The classical SMC and the second order sliding (2-sliding)
mode controllers, including the continuous STW control algorithm,
can robustly handle such problem if the boundaries of the
perturbations are known. The main disadvantage of the classical
SMC is in introducing control chattering, while SOSM controllers
are able to attenuate it. In this work we are looking for an
adaptive-gain STW (ASTW) algorithm that is able to address this
problem via generating continuous control function (chattering
attenuation) so that its gains are adapted to the unknown additive
and multiplicative perturbations with the unknown boundaries
and without the control gain overestimation.

Consider a single-input uncertain nonlinear system

x = f (x) + g (x)u

(1)

where x X Rn is a state vector (X is a compact set), u R

is a control function, f (x) Rn is a differentiable, partially known
vector-field. Assume that

A1. A sliding variable = (x, t ) R is designed so that the

systems (1) desirable compensated dynamics are achieved in
the sliding mode = (x, t ) = 0.
A2. The relative degree of system (1) with the sliding variable
(x, t ) with respect to u equals one, and the internal dynamics
are stable.

2.2. Control structure

The following STW control is considered (Levant, 1993)

= | |1/2 sign( ) + v

v = sign( )

(8)

where the adaptive gains

= ( , , t )
= ( , , t )

(9)

Y. Shtessel et al. / Automatica 48 (2012) 759769

are to be defined. The solution of system (6) is understood in the

sense of Filippov (Filippov, 1988). Assuming the uncertain function
b1 (x, t ) to be an uncertain piece-wise constant, control system (6)
and (8) can be rewritten as

= b1 (x, t )| |1/2 sign( ) + + a1 (x, t )

=

b1 (x, t )

(0) = 0

sign( ) + a 2 (x, t ) + b 1 (x, t )v

(10)

where = a2 + b1 (x, t )v . Next, for brevity, the notations b1 and

a2 will be used for the terms b1 (x, t ) and a2 (x, t ). Let us assume
that the term b 1 v is bounded with unknown boundary 3 > 0, i.e.

|b 1 v|

|b 1 |
2

d 3 .

(11)

It is demonstrated below that the adaptive gain = ( , , t )

is bounded with uncertain boundary > 0, i.e. || . Then
Eq. (11) becomes

|b 1 v|

1
2

|b 1 | t 3 .

(12)

It is worth noting that Eq. (12) is valid on any finite time interval. It
is shown below that the dynamics of (10) are considered on a finite
time interval only. Finally, the boundary of the uncertain function
(
x, t ) = a 2 (x, t ) + b 1 v exists, but is unknown. This is

|(
x, t )| 2 + 3 = 4 .

(13)

The control design problem is reduced to designing ASTW control

(8)(9) that drives , 0 given by (10) in finite time in the
presence of the bounded additive (5), (13) and multiplicative (3)
perturbations with the unknown boundaries 1 , 4 , 1 > 0.
The idea of designing ASTW is to dynamically increase the control gains (t ) and (t ) until the 2-sliding mode establishes. Then
the gains shall start reducing. This gain reduction shall be reversed
as soon as the sliding variable or its derivative start deviating from
the equilibrium point = = 0 in 2-sliding mode. Therefore, a
rule (a detector) that detects the beginning of a destruction of the
sliding mode shall be constructed and incorporated in the ASTW
control law that allows not-overestimating the control gains (t )
and (t ). This detector is proposed to design by introducing a
domain | | that is used as follows: as soon as this domain is
reached the gains (t ) and (t ) start dynamically reducing until
the system trajectories leave the domain. Then the gains start dynamically increasing in order to force the trajectories back to the
domain in finite time.
Remark 1. A selection of the mentioned detector is not unique.
In particular, since the super-twisting algorithm is a 2-SMC
controller, the value of can be also taken into account, while
dynamically adapting the gains of the controller. For instance, the
gains (t ) and (t ) can start decreasing if the condition | | +
c | | 1 or 2 + c 2 2 , c > 0, 1 > 0, 2 > 0, is satisfied.
Such choice of the detector could yield a smaller domain of the
convergence and will be analyzed in a future work. 
2.3. Main results
The main result of the paper is formulated in the following
theorem.

761

x(0), (0), there exist a finite time 0 < tF and a parameter (as soon
as the condition, 4 being defined as 4 = 2 + 3 ,

>

2
21 24 4 2
1 + 4 2 (44 + 1)
+
(1 1 )
12 (1 1 )

holds, if | (0)| > ) so that a real 2-sliding mode, i.e.| | 1

and | | 2 , is established t tF via ASTW control (8) with the
adaptive gains ((0) > m )

sign(| | ),

if > m
if m

= 2
where , , 1 , 1 , are arbitrary positive constants, and 1 ,
2 > 0. The parameter m is an arbitrary small positive constant. 
Proof. The proof is split into two steps. In the first step, we
will present system (10) in a form convenient for the Lyapunov
analysis. In order to do this a new state vector is introduced
z = [z1 z2 ]T = | |1/2 sign( )

(15)

and system (10) can be rewritten as

z1 =

( b1 z1 + z2 + a1 (x, t ))

2|z1 |
b1

z2 =

(16)

2|z1 |

z1 + (x, t )

where (x, t ) = a 2 (x, t ) + b 1 (x, t )v . Eq. (16) can be rewritten in a

vector-matrix format

1
z1
b1
=
z2

b1
2|z1 |

1
0

A(z1 )

1
2|z1 | 0

0
2|z1 |

z1
z2

G(z1 )

a1 (x, t )
.
(x, t )

(17)

Due to Assumption A4 and (13) we can write

a1 (x, t ) = 1 (x, t )| |1/2 sign( ) = 1 (x, t )z1

(x, t ) =

2 (x, t )
2

sign( ) =

2 (x, t ) z1
2
|z1 |

(18)

where 1 (x, y), 2 (x, t ) are some bounded functions so that

0 < 1 (x, t ) 1 ,

0 < 2 (x, t ) 24 .

Eq. (17) can be rewritten in view of Eq. (18)

z1
z
=
A (z1 ) 1
z2
z2

(19)

where
A (z1 ) =

( b1 1 (x, t ))
2|z1 | ( b1 2 (x, t ))
1

1
.
0

It can be observed that

Theorem 1. Consider system (10). Suppose that the functions
a1 (x, t ), a2 (x, t ) and b1 (x, t ) satisfy Assumptions A3 and A5 for
some unknown gains 1 , 2 , 1 > 0. Then, for any initial conditions

(14)

(a) if z1 , z2 0 in finite time then , 0 in finite time;

(b) |z1 | = | |1/2 and sign(z1 ) = sign( ).

762

Y. Shtessel et al. / Automatica 48 (2012) 759769

In the second step of the proof, the stability analysis of system

(17) is performed. In order to do it the following Lyapunov function
candidate is introduced
V (z1 , z2 , , ) = V0 +

21

22

(20)

where (with > 0, > 0)

V0 (z ) = + 4

+ 4 2
P =
2

where z 2 = z12 + z22 = | | + z22 and

(21)

and > 0, > 0 are some constants. It is worth noting that the
matrix P is positive definite if > 0 and are any real number. The
derivative of the Lyapunov function candidate (20) is presented

1

V (z , , ) = z A T (z1 )P + P A (z1 ) z +
1

+
.
2

r =

V (z , , ) = z T Pz + z T P z +

|z1 |

1
2|z1 |

z T Q z .

(23)

Q =

Q 11
Q 21

Q 12
4

Q 11 = 2 b1 + 4 b1 (2 )

2( + 4 2 )1 (x, t ) + 42 (x, t )

= Q 21 = b1 2 b1 4 2 + 21 (x, t ) 2 (x, t ).

x2 + y2 + z 2

enforce

= 2.

(25)

The matrix Q will be positive definite with a minimal eigenvalue

min (Q ) 2 if

In view of (23) and assuming that Eqs. (25)(26) hold, it is easy to

show that
1/2
rV0

V 0

(27)

where

1/2

rV0

1/2

min (P )
r =
.
max (P )

(28)

Indeed, since

2|z1 |

2|z1 |

zT z =

|z1 |

z 2

(29)

1
2
+
| | +
| |. (33)
2
21
22
1

|x| + |y| + |z |

(34)

V (z1 , z2 , , ) 0

V (z1 , z2 , , ) +

2
1
| | +
| |.
+
2
21
22
1

(35)

Now, we assume that the adaptation law (14) makes the adaptive
gains (t ) and (t ) bounded (this assumption will be proven later).
Then there exist positive constants , such that (t ) < 0
and (t ) < 0, t 0. In view of the above assumption,
Eq. (35) can be reduced to the following

| |

1

1
21
1

2

2
22
1

(36)

It gives
V (z1 , z2 , , ) 0 [V (z1 , z2 , , )]1/2 +

(37)

with

1

1
21
1

2

2
22
1

(38)

C1. Suppose that | | > and (t ) > m for all t 0. Then, in

view of (14)

= 1

(39)

and

and

min (P )z 2 z T Pz max (P )z 2

2
1
| |
| | 0 V (z , , )

21
22

= | |

z T Q z

and in view of Eq. (20), we can derive

(26)

V 0 (z )

V (z1 , z2 , , ) 0 [V (z1 , z2 , , )]1/2

2
1 + 4 2 (44 + 1)
21 24 4 2
>
+
.
(1 1 )
12 (1 1 )

1/2

In order to guarantee the positive definiteness of the matrix Q , we

with 0 = min(r , 1 , 2 ). Taking into account Eq. (34), we can

rewrite Eq. (33) as

with

Q 12

Taking into account a well-known inequality

(24)

+
1
2
1
2
1/2
= rV0
| |
| |
21
22

(22)

The symmetric matrix Q is computed taking into account (5) and

(18)

z T Q z +

1/2

rV0

(32)

Now, in view of Eqs. (22) and (27) can be rewritten (with =

The first term of (22) is computed taking into account (17) and (19)

min (P )
.
max (P )

and = )

V 0 = z T A T (z1 )P + P A (z1 ) z

(31)

{P }

1/2

1/2

1/2
min

then
V 0 (z ) rV0 ,

z12 + z22 4 z1 z2 = z T Pz

1/2

V0 ( z )

|z1 | = | |1/2 z

2
= | |

2
22

(30)

(40)

Y. Shtessel et al. / Automatica 48 (2012) 759769

After selecting = 22
1

2
1

and differentiating (25) we obtain

= 2 = 2

2
= 1 21 = 2
.

(41)

In view of Eq. (41) the term in (40) becomes = 0 and

V (z1 , z2 , , ) 0 [V (z1 , z2 , , )]1/2 .

(42)

It is worth noting that for the finite time convergence (t )

must satisfy inequality (26). It means that (t ) shall increase
in accordance with (39) until (26) is met that guarantees the
positive definiteness of the matrix Q and validity of (42). After
that the finite time tF convergence to the domain | | is
guaranteed according to (42).
C2. Suppose that | | < then (t ) is reducing in accordance with
(14) that takes a form

1
2

Note that Eqs. (45) and (46) prove only the existence of the real
sliding mode domain
W = { , : | | 1 , | | 2 , 1 > }

763

if > m

(43)

if m

(48)

since it is practically impossible to calculate the values 1 , 2 .

Theorem 1 is proven. 
Now we can prove the assumption about boundedness of (t )
and (t ).
Proposition 1. The adaptive gains (t ) and (t ) are bounded.

Proof. In the domain < | | 1 , a solution to (14) can be

constructed as

= (0) + 1

1
2

t,

0 t tF .

(49)

Therefore (t ) is bounded. The adaptive gain (t ) is also bounded,

since (t ) = 2(t ). Inside the domain | | the control gains
(t ) and (t ) are decreasing. Therefore, the gains (t ) and (t ) are
bounded in the real 2-sliding mode. Proposition 1 is proven. 
Now we can easily estimate finite reaching time.

and the term

,
2| |
21

=
1

|m + t |

,
1
21

if > m
if m

(44)

becomes (or can be) positive (one recalls that m is a small

parameter). It is worth noting that second equation in (44) is
valid only for finite time, since as soon as becomes less or
equal to m , its value immediately starts increasing such that
= m + t. Then, the first equation in (44) becomes valid.
In view of (44) the derivative of the Lyapunov function (37)
becomes sign indefinite and | | may become larger than due
to decrease of the control gains (t ) and (t ).
As soon as | | becomes greater than the condition that
defines C1 case holds so that reaches the domain | | again
in finite time, and so on. Therefore, during the adaptation process
the sliding variable reaches the domain | | in finite time
then may leave this domain for a finite time, and it is guaranteed
that it always stays in a larger domain | | 1 , 1 > in a real
sliding mode.
Inside the domain | | , the value | | can be estimated in
accordance with (10), (14) and (25)

| | (1 + 1 )

1/2

(t2 ) + 1

+ 1 1 + 4 (t3 t2 ) = 2

(45)

1 1
2

(t3 t2 )
(46)

where t2 and t3 (t3 > t2 ) are the time instants when (t ) leaves the
domain | | and enters this domain afterwards respectfully.
Combining the conditions (45) and (46) we obtain

| | max( 2 , 2 ) = 2 .

2V 1/2 (t0 )

(47)

(50)

where 0 = min(r , 1 , 2 ).

Proof. Inequality (26) is fulfilled in finite time, since its right hand
side is bounded and the adaptive gain (t ) is increasing linearly
with respect to time in accordance with (14). Assuming = 0
implies , 0 in finite time tr that can be estimated by tr
2V 1/2 (t0 )

, which is obtained by a direct integration of inequality (37)

0
with = 0. For > 0, , W ( , ) in finite time tF tr .
Proposition 2 is proven. 
It is worth noting that if the detector for adaptive gain
reduction, the term sign(| | ) in the gain adaptation law (14),
is eliminated (by making = 0), then the adaptive gain law (14)
shall be changed to (with (0) > 0)

where t1 and t2 are the time instants when (t ) enters the domain
| | and leaves this domain respectfully. When | (t )| becomes
< | | 1 then
1/2
1

tF

| | ((1 1 )(t1 ) + 1 ) 1/2

+ [(1 1 )(t1 ) + 4 ] (t2 t1 )
= 2

Proposition 2. As soon as inequality (26) is fulfilled in finite time t0 ,

the adaptive-gain STW control law (14) drives the sliding variable
from the initial condition | (t0 )| and its derivative to the domain
W = { , : | | 1 , | | 2 , 1 > } in finite time that is
estimated as

0,

1
2

if = 0
if = 0

(51)

= 2.
In this case the ASTW control law will drive the systems (1)
trajectory to the ideal 2-sliding mode, i.e. = = 0 in finite
time. However, the adaptive control gains (t ) and (t ) can be
overestimated. This result is formulated in the following corollary.
Corollary 1. Consider system (10). Suppose that the functions
a1 (x, t ), a2 (x, t ) and b1 (x, t ) satisfy Assumptions A3 and A5 for
some unknown gains 1 , 2 , 1 > 0. Then for any initial conditions
x(0), (0), the second order sliding mode, i.e. = = 0, will
be established in finite time via STW control (8) with the adaptive
gains (51). 
The Propositions 1 and 2 are obviously valid for the case of
ASTW control in (8) and (51).

764

Y. Shtessel et al. / Automatica 48 (2012) 759769

Matlab/Simulink allows to use a DS1104 board (dSpace Co.) on

which the control law is implemented. In the sequel, the experimental results have been obtained with a 1 ms sample time.
3.2. Simulation model

Fig. 1. Photo of the electropneumatic systemOn the left hand side is the
main actuator whose its position is controlled. On the right hand side is the
perturbation actuator whose the load force is controlled.

A standard pneumatic actuator is equipped by a pneumatic

damper in order to protect the piston: this protection avoids
high clashes between the piston and the external structure of the
actuator. The damper is composed by a restriction which limits the
exhaust mass flow rate. In order to obtain maximum performance,
this restriction has been deleted. It implies that, in a first step, the
control law has to be evaluated on cosimulation. The cosimulation
is using the simulation model developed with AMESim software,
this model trying to be as close as possible to the physical behavior.
Servodistributor model. The servodistributor model is composed
in two parts, a dynamic part and a static one:

Dynamic part is modelized by a second order transfer function

3. Electropneumatic system

identified from experimental measure

3.1. Description

F(s) =

The electropneumatic system (see Figs. 12) is composed by

two actuators. The first one, named the main one (left hand
side), is a double acting electropneumatic actuator controlled by
two servodistributors (Fig. 2) and is composed by two chambers
denoted P and N. Piston diameter is 80 mm and rod diameter
25 mm. With a source pressure equal to 7 bar, the maximum
force developed by the actuator is 2720 N. The air mass flow
rates qm entering in the chambers are modulated by two threeway servodistributors. The pneumatic jack horizontally moves a
load carriage of mass M. This carriage is coupled to the second
electropneumatic actuator, the so-called perturbation one. As
previously mentioned, the goal of this latter is to produce a
dynamical load force on the main actuator. The actuator has the
same mechanical characteristics than the main one, but the air
mass flow rate is modulated by a single five-way servodistributor.
In the sequel of the paper, only the control of the main
actuator position is considered; note that the force control of
the perturbation actuator is currently made by an analogic PID
controller developed by the test bench constructor. In conclusion,
the aim of this test bench is to evaluate performances of position
controller with respect to unknown dynamical perturbation force.
The experimental test bench is simulated with a fluid power
systems dedicated software AMESim (LMS SA Co.), and the
control law is developed under Matlab/Simulink (The Mathworks
Co.). It implies a cosimulation program (Figs. 3 and 4) through
links between AMESim and Matlab/Simulink. In Fig. 4, the block
AMESim Model makes the link between Matlab simulation
and Amesim simulator described by Fig. 3. A consequence of
the cosimulation is that two models are used: a simulation
model simulated by AMESim, and a control model simulated by
Matlab/Simulink. It can be summarized as follows

the simulation model takes into account physical phenomena

as temperature variations, experimental values of mass flow
rate delivered by each servodistributors, dynamics of servodistributors, dry friction..., and is developed under Amesim. The
perturbation force is viewed as an input.
the control model is simpler than the previous (for example,
mass flow rates models are written as pressures polynomials
(Sesmat & Scavarda, 1996)) and issued in order to design
the nonlinear position controller under Matlab/Simulink. The
perturbation force is supposed unknown.

2
ns
2
s2 + 2 s ns s + ns

(52)

with ns = 246 rad s1 and s = 0.707.

Static part is modelized by an experimental table in which
mass flow rate is given in function of ratio pressure (upstream/
downstream) and control voltage (Sesmat & Scavarda, 1996).
Pneumatic chamber variable volume model. Each chamber of
the pneumatic actuator is considered as a variable volume, in
which the air mass evolves with time. State the classical following
assumptions (Shearer, 1956):

H1. Air is perfect gas and its kinetic inconsequential.

H2. The pressure and the temperature are homogeneous in each
chamber.

H3. The mass flow is pseudo-stationary.

The first dynamic principle applied to the air mass and the
thermodynamic evolution of air in each chamber read as (with X =
P or N) (Shearer, 1956)
dpX
dt

=
+

dTX
dt

pX dVX

rTr

VX dt
VX
( 1) dQX
VX

= ( 1)

rTX2
p X VX

dt
TX dVX
VX dt

qmXin

rTX
pX VX

rTX
VX

qmXout

(53)

( Tr TX )qmXin

( 1)qmXout + ( 1)

TX

dQX

pX VX dt

with the adiabatic constant, Tr the temperature inside the

upstream tank, qmXin the mass flow rate brought inside the X
chamber, and qmXout the mass flow rate brought outside the X
chamber. QX , the thermal exchange with the X chamber wall, is
described by Assumption A4.

H4. The thermal exchange is due only by conduction described

by
dQX
dt

= ScX (TcX TX )

(54)

with the thermal exchange coefficient by conduction, ScX the

total area inside a X chamber, and TcX the temperature of the X
chamber wall.

Y. Shtessel et al. / Automatica 48 (2012) 759769

765

Fig. 2. Scheme of electropneumatic systemThis figure displays the mechanical and software structures. The software structure is based on a dSpace board on which the
position controller of the main actuator is implemented. The mechanical structure is composed by two actuators, the main one (left hand side) and the perturbation
one (right hand side).

Fig. 3. AMESim model for cosimulation.

Fig. 4. MATLAB/Simulink control law for cosimulation.

766

Y. Shtessel et al. / Automatica 48 (2012) 759769

With VP (y) = V0 + S y and VN (y) = V0 S y (V0 being equal

to the half of the cylinder volume), the model used for the design
of controller is a nonlinear system and reads as

Mechanical model. The second Newton law gives

dv

dt
dy

1
S (pP pN ) Ff bv v F
M

(55)

with friction force Ff including stiction, Coulomb and Stribeck

phenomena.
Samplers and saturation. Samplers are added in AMESims model
in order to take into account samplers of acquisition card; sample
time is 1 ms which is very smaller than the natural frequency of
this electropneumatic system. So it is not necessary to discretize
the model all the control law are synthesize in continuous time.
Saturation signal control are added, i.e. |usat | = 10 V.
3.3. Control model
This model is developed in order to design the control law
in order to obtain a simplest version which allows the design of
control law. The following hypotheses are added

H5. The process is polytropic and characterized by coefficient k

(with 1 < k < ).
H6. The leakage between system and atmosphere are neglected.
H7. The temperature variations in each chamber are inconsequential with respect to the supply temperature, i.e. TP =
TN = T .

dpX
dt

= k

pX dVX
VX dt

krT
VX

qmXin qmXout .

(56)

H8. The leakages between the two chambered, and between the
servodistributor and the jack are negligible.
H9. Supply and exhaust pressure are supposed to be constant.
By defining qm (uX , pX ) := qmXin qmXout , one gets
dpP
dt
dpN
dt

= k
= k

pP

dVP (y)

VP (y) dt
pN dVN (y)
VN (y)

dt

krT

qm (uP , pP )
VP
krT
+
qm (uN , pN ).
VN

v =

VP (y)

1
M

(57)

H10. All dry frictions forces are neglected.1

H11. There is no control signal saturation.
H12. Dynamic part of servodistributor is neglected.2 .
H13. Static part of servodistributor depends on pressures and
control value
qm (uX , pX ) = (pX ) + (pX , sign (uX )) uX

with and 5th-order polynomials with respect to pX (Sesmat

& Scavarda, 1996) and issued from experimental measures.
H14. Only the position of the actuator is controlled, which
means that the problem is a single inputsingle output (SISO).
It implies that uP = uN = u.

(58)

[S (pP pN ) bv v F ]

y = v
with F the unknown perturbation force, P = (pP ), N = (pN ),

P = (pP , sign (u)) ,

N = (pN , sign (u)) .
It is obvious that the system (58) reads as the nonlinear system
(1) with x = [pP pN v y]T ,

S
P pP v
VP (y)
rT

krT
S
N + pN v
VN (y)
rT
krT

,
f ( x) =

[S (pP pN ) bv v F ]
M

g (x) =

Therefore, pressures dynamics reads as

P + P u pP v
rT

krT
S
p N =
N N u + pN v
VN (y)
rT
p P =

=v

dt

krT

krT
VP (y)

v
P

krT
VN (y)

N 0 0

3.4. Control problem formulation

In system (58), the vector-fields f (x) and g (x) are partially
known. Furthermore, the perturbation terms F and bv v are
unknown and bounded with unknown bounds, whereas the
functions P , N , P and N give only an estimation of the mass
flow rate, the quality of this estimation being evaluated with
difficulty. The lack of knowledge of the perturbation/uncertainties
bounds makes it difficult designing a robust sliding mode
controller without overestimating the control gains. The problem
is then to design a robust continuous adaptive-gain super-twisting
sliding mode controller u that drives the output (the position y)
of the electropneumatic actuator to follow a prescribed profile
(see Fig. 5dotted line) in the presence of bounded external
perturbations (see external forces F Fig. 9) with unknown
bounds.
4. ASTW sliding mode control of an electropneumatic actuator:
design and experimental results
The aim of the control law is to get a good accuracy in term
of position tracking for the desired trajectory displayed in Fig. 5
(dotted line) in spite of the parametric uncertainties given above;
furthermore, the dry frictions have not been taken into account in
the controller design. Following the previous section, consider the
sliding variable

= n2 (y yd (t )) + 2 n (y y d (t )) + (y y d (t ))

(59)

with = 0.7 and n = 50 rad s . As system (58)(59) admits a

relative degree equal to 1 for with respect to u, one gets
1

1 The viscous friction forces have been identified on real system: it has been
established that the carriage presents such frictions bv v with bv = 30.
2 This hypothesis has been made given that the servodistributor dynamics are
much faster than the mechanical part of the system and are considered as singular
perturbed unmodeled dynamics (Fridman, 2003; Soto-Cota, Fridman, Loukianov, &
Canedo, 2006)

= a() + b() u.

(60)

From Laghrouche, Smaoui, Brun, and Plestan (2004), functions a

and b are bounded for x X. Furthermore, the function b fulfills

Y. Shtessel et al. / Automatica 48 (2012) 759769

Fig. 5. External force1000 N. Measured (solid line) and desired (dotted line)
actuator position y (m) versus time (s).

767

Fig. 6. External force1000 N. Position tracking error (m) versus time (s).

Assumption A3 for x X. Furthermore, the function a can be

written as a = a0 + a. It yields that the control law reads as
u=

1
b0

(a0 + )

(61)

with control input defined by (8)

= | |1/2 sign( ) + v

v = sign( ),

(62)

the gains and being defined by (14)

= 1 2 sign(| | ),

Fig. 7. External force1000 N. Gain (t ) versus time (s).

if > m
if m

(63)

= 2.
Parameters of the controller have been tuned as (this tuning
has been made in order to get the best behavior and high
performances)

= 1,
1 = 2,
1 = 200,
m = 0.01,
= m .

= 0.7,

Several experimental tests have been made in order to highlight

the properties of this class of controllers: then, experimentations
have been with two different magnitudes of perturbations in order
to show the adaptation of the gain with respect to it. Secondly,
experimental tests have been with a constant gain, whose the value
has been sufficiently large in order to counteract the worst case of
uncertainties/perturbations: it yields to an overestimation of the
gain.
Figs. 56 display desired and current positions, and the position
tracking error for an external 1000 N force, respectively, and show
the effectiveness of the adaptive-gain STW controller given that
the trajectory tracking is accurate in spite of bounded additive and
multiplicative uncertainties with unknown bounds.
The time history of the adaptive gain (t ) is shown in Fig. 7,
where the reduction of the gain during is demonstrated whereas
the tracking is accurate in spite of the external force F (see Fig. 9).
With a lower magnitude of the perturbation force (500 N), Fig. 11
shows that the gain is lower: there is really an adaptation of the
gain magnitude with respect to the perturbation magnitude. Fig. 10
also shows that the tracking error is less and has smaller transients.
Fig. 8 displays the control input which is far from saturation.
Note that, in case of non adaptive gain, the input is larger: in fact,
if all the uncertainties/perturbations are taken into account in the
worst case, the gain has to be strongly over-estimated and has
been tuned at 4000. Fig. 12 displays the control input: it is clear
that the control magnitude is larger.

Fig. 8. External force1000 N. Control input u(t ) (V ) versus time (s).

Fig. 9. External perturbation force (N) versus time (s).

5. Conclusions
A novel adaptive-gain real super-twisting (ASTW) sliding
mode controller is proposed. The both drift uncertain term
and multiplicative perturbation are assumed to be bounded
with unknown boundaries. The proposed Lyapunov-based ASTW
controller design dynamically adapted control gain that ensures

768

Y. Shtessel et al. / Automatica 48 (2012) 759769

Fig. 10. External force500 N. Position tracking error (m) versus time (s).

Fig. 11. External force500 N. Gain (t ) versus time (s).

Fig. 12. External force1000 N. Constant gain = 4000. Control input u(t ) (V )
versus time (s).

the establishment, in a finite time, of a real second order sliding

(2-sliding) mode. The proposed ASTW sliding mode control
does not overestimate the values of the control gains. Finite
convergence time is estimated. The efficacy of the proposed ASTW
sliding mode control algorithm is confirmed via its application to
position control of an electropneumatic actuator.
References
Boiko, I. (2008). Discontinuous systems. Boston, USA: TBirkhauser.
Bouri, M., & Thomasset, D. (2001). Sliding control of an electropneumatic actuator
using an integral switching surface. IEEE Transactions on Control Systems
Technology, 2(2), 368375.
Brun, X., Sesmat, S., Thomasset, D., & Scavarda, S. (1999). A comparative study
between two control laws of an electopneumatic actuator. In Proceedings of the
european control conference ECC99. Karlsruhe. Germany.
Brun, X., Thomasset, D., & Bideaux, E. (2002). Influence of the process design on
the control strategy: application in electropneumatic field. Control Engineering
Practice, 10(7), 727735.
Burton, J. A., & Zinober, A. S. I. (1986). Continuous approximation of vsc. International
Journal of Systems Sciences, 17, 875885.
Chiang, M. H., Chen, C. C., & Tsou, T. N. (2005). Large stroke and high precision
pneumatic-piezoelectric hybrid positioning control using adaptive discrete
variable structure control. Mechatronics, 15(5), 523545.

Djema, M., Barbot, J. P., & Busawon, K. (2008). Designing r-sliding mode control
using smooth iterative manifolds. Medical Journal of Measurement and Control,
4(2), 8693.
Edge, K. A. (1997). The control of fluid power systemsresponding to the challenge.
Journal of Systems and Control Engineering, 211(2), 91110.
Edwards, C., & Spurgeon, S. (1998). Sliding mode control: theory and applications.
Bristol, UK: Taylor & Francis.
Filippov, A. F. (1988). Differential equations with discontinuous right-hand side. The
Netherlands: Kluwer, Dordrecht.
Fridman, L. (2001). An averaging approach to chattering. IEEE Transaction on
Automatic Control, 46(8), 12601265.
Fridman, L. (2002). Singularly perturbed analysis of chattering in relay control
systems. IEEE Transaction on Automatic Control, 47(12), 20792084.
Fridman, L. (2003). Chattering analysis in sliding mode systems with inertial
sensors. International Journal of Control, 76(9/10), 906912.
Girin, A., & Plestan, F. (2009). A new experimental setup for a high performance
double electropneumatic actuators system. In Proceedings of the american
control conference 2009. Saint-Louis, Missouri, USA.
Hamiti, K., Voda-Besanon, A., & Roux-Buisson, H. (1996). Position control of a
pneumatic actuator under the influence of stiction. Control Engineering Practice,
4(8), 10791088.
Kimura, T., Hara, S., Fujita, T., & Kagawa, T. (1997). Feedback linearization for
pneumatic actuator systems with static friction. Control Engineering Practice,
5(10), 13851394.
Kyoungkwan, A., & Shinichi, Y. (2005). Intelligent switching control of pneumatic
actuator using on/off solenoid valves. Mechatronics, 15, 683702.
Laghrouche, S., Plestan, F., & Glumineau, A. (2007). Higher order sliding mode
control based on integral sliding surface. Automatica, 43(3), 531537.
Laghrouche, S., Smaoui, M., Brun, X., & Plestan, F. (2004). Second order sliding mode
controller for electropneumatic actuators. In Proceedings of the american control
conference 2004. Boston, Massachusetts.
Laghrouche, S., Smaoui, M., Plestan, F., & Brun, X. (2006). Higher order sliding
mode control based on optimal approach of an electropneumatic actuator.
International Journal of Control, 79, 119131.
Levant, A. (2003). Higher-order sliding modes, differentiation and output-feedback
control. International Journal of Control, 76(910), 924941.
Levant, A. (2005). Homogeneity approach to high-order sliding mode design.
Automatica, 41(5), 823830.
Levant, A. (1993). Sliding order and sliding accuracy in sliding mode control.
International Journal of Control, 58, 12471263.
Ming-Chang, S., & Shy-I, T. (1995). Identification and position control of a servo
pneumatic cylinder. Control Engineering Practice, 3(9), 12851290.
Miyajima, T., Fujita, T., Sakaki, K., Kawashima, K., & Kagawa, T. (2007). Development
of a digital control system for high-performance pneumatic servo valve.
Precision Engineering, 31, 156161.
Moreno, J.A., & Osorio, M. (2008). A lyapunov approach to second order sliding mode
controller and observers. In Proceedings of the IEEE conference on decision and
control. Cancun, Mexico.
Paul, A. K., Mishra, J. K., & Radke, M. G. (1994). Reduced order sliding mode control
for pneumatic actuator. IEEE Transactions on Control Systems Technology, 2(3),
271276.
Plestan, F., Glumineau, A., & Laghrouche, S. (2008). A new algorithm for high
order sliding mode control. International Journal of Robust and Nonlinear Control,
18(45), 441453.
Plestan, F., Moulay, E., & Glumineau, A. (2010a). Output feedback sampling control:
a robust solution based on second order sliding mode. Automatica, 46(6),
10961100.
Plestan, F., Shtessel, Y., Brgeault, V., & Poznyak, A. (2010b). New methodologies
for adaptive sliding mode control. International Journal of Control, 83(9),
19071919.
Polyakov, A., & Poznyak, A. (2009). Reaching time estimation for super-twisting
second order sliding mode controller via lyapunov function designing. IEEE
Transactions on Automatic Control, 54(8), 19511955.
Rao, Z., & Bone, G.M. (2006). Modeling and control of a miniature servo pneumatic
actuator. In Proceedings of the IEEE international conference on robotics and
automation ICRA06. Orlando, Florida, USA.
Richard, E., & Scavarda, S. (1996). Comparison between linear and nonlinear control
of an electropneumatic servodrive. Journal of Dynamic Systems, Measurement,
and Control, 118, 245252.
Schultea, H., & Hahn, H. (2004). Fuzzy state feedback gain scheduling control of
servo-pneumatic actuators. Control Engineering Practice, 12(5), 639650.
Sesmat, S., & Scavarda, S. (1996). Static characteristics of a three way servovalve. In
Proceedings of the conference on fluid power technology. Aachen, Germany.
Shearer, J. L. (1956). Study of pneumatic process in the continuous control of
motion with compressed air. Transactions of the American Society of Mechanical
Engineers, 78, 233249.
Shtessel, Y.B., Moreno, J.A., Plestan, F., Fridman, L.M., & Poznyak, A.S. (2010). Supertwisting adaptive sliding mode control: a lyapunov design. In Proceedings of the
IEEE conference on decision and control CDC 2010. Atlanta, Georgia, USA.
Shtessel, Y., Shkolnikov, I., & Levant, A. (2007). Smooth second order sliding modes:
missile guidance application. Automatica, 43(8), 14701476.
Slotine, J. J. E., & Li, W. (1991). Applied nonlinear control. Englewood Cliffs, New
Jersey, USA: Prentice Hall.
Smaoui, M., Brun, X., & Thomasset, D. (2006). A study on tracking position control
of electropneumatic system using backstepping design. Control Engineering
Practice, 14(8), 923933.
Smaoui, M., Brun, X., & Thomasset, D. (2005). A combined first and second order
sliding mode approach for position and pressure control of an electropneumatic
system. In Proceedings of the american control conference ACC05. Portland,
Oregon, USA.

Y. Shtessel et al. / Automatica 48 (2012) 759769

Soto-Cota, A., Fridman, L., Loukianov, A., & Canedo, J. (2006). Variable structure
control of synchronous generator: singularly perturbed analysis. International
Journal of Control, 79(1), 113.
Utkin, V., Guldner, J., & Shi, J. (1999). Sliding modes in electromechanical systems.
Bristol, UK: Taylor & Francis.
Yang, L., & Lilly, J.H. (2003). Sliding mode tracking for pneumatic muscle actuators in
bicep/tricep pair configuration. In Proceedings of the american control conference
ACC03. Denver, Colorado, USA.

Yuri Shtessel received the M.S. and Ph.D. degrees in Electrical Engineering with focus on Automatic Control from
the Chelyabinsk State Technical University, Chelyabinsk,
Russia in 1971 and 1978, respectively. Since 1993, he
has been with the Electrical and Computer Engineering Department, The University of Alabama in Huntsville,
where his present position is Professor. His research interests include sliding mode control and observation with
applications to electric power system, aerospace vehicle
control and blood glucose regulation. He published more
than 300 technical papers. Dr. Shtessel is a recipient of the
IEEE Third Millennium Medal for the outstanding contributions to control systems
engineering, 2000. He holds the ranks of Associate Fellow of AIAA and Senior Member of IEEE.

769
Mohammed Taleb received his Diploma in Engineering
in Electromechanical Engineering from ENSAM, Moulay
Ismail University, Meknes, Morocco in 2003, and his
Masters degree in automatic control from the Ecole
Centrale de Nantes, Nantes, France, in 2010. He is currently
working toward the Ph.D. degree in the Institut de
Recherche en Communication et Cyberntique de Nantes
(IRCCyN), Ecole Centrale de Nantes, France. His research
interests include robust nonlinear control and adaptive
higher order sliding mode and their applications to
electropneumatic actuators.

Franck Plestan received the Ph.D. in Automatic Control

from the Ecole Centrale de Nantes, France, in 1995. From
September 1996 to August 2000, he was with Louis
Pasteur University, Strasbourg, France. In September 2000,
he joined the Ecole Centrale de Nantes, Nantes, France
where he is currently Professor. His research interests
include robust nonlinear control (adaptive/higher order
sliding mode), theoretical aspects of nonlinear observer
design and control of electromechanical and mechanical
systems (pneumatic actuators, biped robots, electrical
motors).