1082 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO.

5, MAY 2009

A Sub-Optimal Second Order Sliding Mode Controller order derivative of the considered sliding variable, forcing to zero both
for Systems With Saturating Actuators the sliding variable and a certain number of its time derivatives. The
resulting control variable which acts on the first-order time derivative
Antonella Ferrara, Senior Member, IEEE, and of the sliding variable is continuous, because it is obtained integrating
Matteo Rubagotti, Student Member, IEEE one or more times the discontinuous signal generated by the controller.
In particular, in this technical note, a second order sliding mode
(SOSM) controller [13], designed according to the so-called ‘sub-op-
Abstract—In this technical note, the problem of the possible saturation
timal’ approach [14]–[16] is taken into account. This controller acts
of the continuous control variable in the sub-optimal second order sliding discontinuously on the second-order time derivative of the sliding vari-
mode controller, applied to systems with saturating actuators, is addressed. able, so that the control variable which acts on the system is the inte-
It is proved that during the sliding phase, if basic assumptions are made, the gral of this discontinuous control. If the continuous control signal is
continuous control variable never saturates. On the contrary, during the affected by a saturation, the behavior of the system can be different
reaching phase, the presence of saturating actuators can make the steering
of the sliding variable to zero in a finite time not always guaranteed. In from that expected, and even the stabilization of the sliding variable
the present technical note, the original algorithm is modified in order to to zero cannot be assured any more. To solve the problem, a desatura-
solve this problem: a new strategy is proposed, which proves to be able to tion strategy is proposed, which produces appreciable results, allowing
steer the sliding variable to zero in a finite time in spite of the presence of to maintain the finite time convergence to zero of the sliding variable
saturating actuators.
and the consequent attainment of the control objective in spite of the
Index Terms—Saturation, sliding mode control, uncertain systems. saturating actuators. The problem of saturation of the continuous con-
trol variable has already been faced for other SOMS algorithms, which
have been adapted for saturation in [11] and [12].
I. INTRODUCTION A preliminary version of the theoretical development, describing
Saturation is a particular kind of nonlinearity which affects many dy- only a particular case of the control strategy here proposed, can be
namical systems [1]. In particular, when a controller is designed, one found in [17].
must take into account that the generated control signal can be in fact The technical note is organized as follows: Section II is devoted to
upper and lower-bounded by the actuator saturation. If the controller the formulation of the control problem. Section III analyzes in which
design does not take into account explicitly the presence of such a con- cases a saturation of the continuous control signal can occur and which
straint on the control variable, if the bounds are reached, the actual effects the saturation can provoke on the controlled system when the
control signal acting on the system is different from the one gener- sub-optimal SOSM control algorithm is applied. In Section IV a mod-
ated by the controller. A very common case when this happens is the ified sub-optimal SOSM controller capable of solving the problem in
so-called ‘wind-up’ effect that affects PID controllers [2], and, in gen- question is proposed and analyzed. Simulation results are shown and
eral, any controller which takes into account the integral of the error discussed in Section V, while some conclusions are gathered in Sec-
signal. Strategies to solve this problem have been proposed, in order to tion VI.
avoid undesired behaviors (typically, longer settling times and higher
overshoots [2]). To take explicitly into account the constraints on the II. PROBLEM FORMULATION
control variable amplitude is not easy in general, because saturation is a
Consider a class of uncertain nonlinear system, described by
nonlinear effect. Nevertheless, there exist control methodologies which
succeed in doing this, such as model predictive control [3], where a
constrained optimization problem is solved on-line. This characteristic
x_ (t) =  (x(t); t) + (x(t); t) u(t) (1)
has contributed significantly to the success in industrial applications of
such methodologies.
where x 2 X  IRn is the state vector, while
Sliding mode control is a particular kind of robust control (see
[4], [5] or [6]), which allows the controller to completely reject any
disturbance acting on the control channel (i.e., the so-called ‘matched u 2 [0 U ; U ] (2)
disturbances’, which may also include parameter uncertainties) when
the ‘sliding variable’ (i.e the variable which is directly steered to is the control variable,  : IRn+1 ! IRn and : IRn+1 ! IRn are
zero in a finite time by the controller action) is equal to zero, and the uncertain and sufficiently smooth vector fields. The control objective
so-called ‘sliding mode’ is enforced. In conventional sliding mode is to steer to zero a scalar output variable, called ‘sliding variable’,
control, the generated control signal is discontinuous, which can defined as 1 = 1 (x(t); t). This variable has relative degree r = 1
generate the so-called ‘chattering’ effect, that is, a very high frequency with respect to the control variable u(t). Moreover, a diffeomorphism
oscillation of the sliding variable around zero (see [7]–[10]). To reduce : IR
n01 2 IR ! IRn is defined, such that
this undesired effect, many solutions have been proposed. Among
them, one of the most promising is based on the generation of “higher
order sliding modes” (see, for instance, [11] and [12]). When such a x(t) =  t ; 1 (t))
( ( ) (3)
solution is used, the control variable acts discontinuously on a high
where  (t) 2 IRn01 is the internal state assumed to be BIBO
(Bounded-Input Bounded-Output) stable. System (1) can be trans-
Manuscript received November 09, 2007; revised July 01, 2008. Current ver- formed into the normal form
sion published May 13, 2009. Recommended by Associate Editor F. Wu.
The authors are with the Department of Computer Engineering and Systems
Science, University of Pavia, Pavia 27100, Italy (e-mail: antonella.fer-
[email protected]; [email protected]).
_ 1 (t) = f ( (t); 1 (t); t) + g ( (t); 1 (t); t) u(t) (4)
Digital Object Identifier 10.1109/TAC.2008.2010992 _(t) = ( (t); 1 (t); t) (5)

0018-9286/$25.00 © 2009 IEEE

Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 5, MAY 2009 1083

where f ( (t); 1 (t); t) 2 IR and g ( (t); 1 (t); t) 2 IR are uncertain

and sufficiently smooth functions, while ( (t); 1 (t); t) 2 IRn01 is
a known and sufficiently smooth vector field. Because of the existance
of the diffeomorphism in (3), it is possible to express the functions
which depend on  (t) and 1 (t) as functions of x(t). Now, assume
that the first order input-output dynamics (4) is globally bounded, that
the following bounds are defined:
Fig. 1. Scheme of the control system using conventional sub-optimal SOSM
controller.
jf (x(t); t)j  F (6)
0 < G1  g (x(t); t)  G2 (7)
where, if (1 (t) 0 0:5M )(M 0 1 (t)) > 0 then = 3 2 (0; 1] \
(0; 3G1 =G2 ), else = 1, while M is the last extremal value of 1 ,
where F , G1 and G2 are positive scalars, and that U in (2) is such that set equal to 1 (t0 ) at the starting time instant, and
H 4H
W > max 3 G1 ; 3G1 : (15)
U >
F
: (8)
0 3 G2
G1
In the case without saturating actuator, this control law solves the
Given system (4), (5) with the bounds (2), (6)–(8), the problem dealt problem of steering both 1 (t) and 2 (t) to zero in a finite time,
with in this technical note is to design a continuous control law capable provided that the unknown disturbances are such that (7) holds and,
of making the system state evolve from a certain finite time instant in addition
tr onward on the manifold 1 (t) = 0. When 1 (t) = 0 the system
exhibits the so-called sliding mode, in which the internal state evolves jh (x(t); u(t); t)j  H (16)
according to
where H is a positive scalar.
_(t) = ( (t); 0; t) : (9) In the conventional design of the sub-optimal second order sliding
mode control law the existence of the bound (2) on the maximum am-
The properties of this motion depend on the choice of the sliding vari- plitude of the actual control signal is not taken into account. The ac-
able 1 (t). Note that assumption (8) implies that a first order control tual control u(t) is obtained by integrating the discontinuous auxiliary
law control signal w(t) and, a-priori, it is not possible to assess if u(t) will
exceed the saturation bounds [0U; U ]. Note that condition (2) is useful
u(t) = 0U sign (1 (t)) (10) to assure the boundedness of (16).

is capable of making 1 (t) = 0 in a finite time. Yet, this is a discon-

III. SATURATION PROBLEM
tinuous control law, which can be a limit to its applicability in many
engineering contexts, such as those of mechanical type. If the relative Now consider the case in which the sub-optimal algorithm is used in
degree of the system with respect to the control variable is equal to a control system with a saturating actuator. In particular, assume that
one, like in system (4), second order sliding mode control can be used the plant model belongs to the class described in Section II. As previ-
to avoid the application of a discontinuous control action. Indeed, a new ously mentioned, since the system relative degree is one, the actual con-
t
control variable can be defined as trol signal u(t) is continuous and given by u(t) = u(t0 )+ t w( )d .
The scheme of the control system is reported in Fig. 1: the sub-op-
w(t) = u_ (t) (11) timal controller has the sliding variable 1 (t) (which is regarded as the
output of the plant) as input signal, and generates, as output, the signal
which directly acts on  1 (t). The design of the controller can be carried w(t), which is then integrated. The dashed line wraps all the blocks
on so that the auxiliary control signal w(t) turns out to be discontin- that are part of the controller. The control signal u(t) is then fed into
uous while u(t), the actual control, is continuous. More precisely, by the process (grey rectangle) as input; the process is characterized by
differentiating with respect to time (4), the following auxiliary system the presence of a saturation on u(t), which generates the signal uS (t)
is obtained, the state of which consists of the sliding variable and its with 0U and U as lower and upper-bounds, respectively. The saturated
first time derivative, i.e. signal is the actual input to the plant.
To analyze the effects of the saturation of the control variable u(t),
let us take into account two different phases: the reaching phase and
_ 1 (t) = f (x(t); t) + g (x(t); t) u(t) the sliding phase. During the reaching phase the controller is expected
= 2 (t) (12) to steer 1 (t) and 2 (t) from their initial values to zero in a finite time;
then, during the sliding phase, it is required that the auxiliary system
_ 2 (t) = f_ (x(t); t) + g_ (x(t); t) u(t) + g (x(t); t) u_ (t) state is kept to the origin, so that the controlled system evolves in sliding
= h (x(t); u(t); t) + g (x(t); t) w (t) (13) mode. The sliding phase, by virtue of (8), will be shown to be insen-
sitive to the presence of the saturation of u(t), and for this reason it is
analyzed first; then, the problems which can occur during the reaching
where h(x(t); u(t); t) = f_(x(t); t) + g_ (x(t); t)u(t). Now, design the
phase are briefly discussed.
auxiliary control law relying on the classical “sub-optimal” control ap-
proach [14], [15]
A. Sliding Phase
Once the system is in sliding mode, the following result can be
w(t) = 0 W sign [1 (t) 0 0:5M ] (14) proved.
Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
1084 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 5, MAY 2009

Fig. 2. Time evolutions of the actual control signal u (t) and of u(t). Fig. 3. Trajectory of the auxiliary system state in the considered example.

Theorem 1: Consider the auxiliary system (12), (13), with bounds auxiliary system state: this latter never reaches the origin of the 1 0 2
(6)–(8) and (16), and w(t) as in (14). If 1 (t) = 2 (t) = 0 for all plane.
t  tr , tr being the time instant when the sliding manifold is reached,
then IV. DESATURATION STRATEGY

ju(t)j < U 8t  tr : (17)

To face the problem previously discussed, a modification of the
sub-optimal SOSM controller is hereafter proposed. The obtained
Proof: The actual control signal u(t) is the time integral of a controller will be called in the reminder of the technical note “modified
bounded function w(t), and so it is a continuous function. The func- sub-optimal controller.” The control law (14) is modified as follows:
tions f (x(t); t) and g (x(t); t) are continuous by assumption, and so is
their ratio, being g (x(t); t) 6= 0. Since for t > tr , 2 (t) = f (x(t); t)+ 0 W sign (1 (t) 0 M ) if ju(t)j < U
g(x(t); t)u(t) = 0, then one has that w(t) = 0W sign (u(t)) if ju(t)j  U
(20)

u(t) = 0 fg ((xx((tt));; tt)) : (18) where 2 [0:5; ], and  is equal to 0.5 at the initial time instant t0 . It
is important to note that the value of at each iteration can be chosen
Taking into account the bounds (6)–(8), it follows that as any value in the specified interval. Then, if, at a time instant ts ,
ju(ts )j  U , then the value of  is immediately changed to
ju(t)j = jjfg ((xx((tt));; tt))jj  GF1 < U (19)
= 1 (ts )
M (21)
which concludes the proof.
Remark 1: The focus of Theorem 1 is on the fact that the presence of where M is the last extremal value of 1 (t). Finally, when a new
the actuator saturation does not affect the performance while in sliding extremal value of 1 (t) is reached, the value of  is re-initialized to
mode. Yet, one can also note that, in case one between f (x(t); t) or 0.5. The variable is defined similarly to in the “classical” sub-optimal
g(x(t); t) is a known quantity, the availability of a measurement of u(t) algorithm: if (1 (t) 0 M )(M 0 1 (t)) > 0 then = 3 2 (0; 1] \
permits to identify the other quantity, which is a-priori unknown. Note (0; 3G1 =G2 ), else = 1. Note that, if u(t) is not directly measurable
also that, while in the sliding phase, u(t) coincides with the so-called but the value of U is known, the saturation due to the actuator can be
equivalent control defined in [4] for the conventional first order sliding replicated in the modified sub-optimal controller, as it happens in anti
mode case. wind-up schemes.
Remark 2: This result is independent of the use of the sub-optimal Remark 3: It is worth noting that the second equation in (20) means
controller, since the only property of the controller that has been ex- that, when u(t) reaches one of its saturation values, its value is forced to
ploited is the capability of keeping 2 (t) to zero after a finite period of decrease. Anyway, if the switching value of 1 (t) has not been reached
time, which is true for any second order sliding mode controller. yet, then u(t) increases again (due to the use of the first equation in
(20)). As a result, u(t) remains at the saturation value until the new
B. The Reaching Phase switching value is reached.
Remark 4: Using the described strategy, when a saturation occurs,
During the reaching phase, since w(t) can be constant for a relatively the controller, in practice, can make u(t) start to decrease its absolute
long time interval, u(t) can reach one of the saturation bounds. From value at any time instant, provided that such instant comes before tl ,
this moment on, the sequence of the extremal values of variable 1 (t) in tl being the time instant when 1 (tl ) = 0:5M . In the following the-
the saturated case differs from the one attainable when the saturation is orem, it will be proved that, for any chosen switching time instant in
not present. This implies that, in principle, relying on the assumptions the allowed interval, the proposed modified sub-optimal control law
of the sub-optimal algorithm, the convergence to the origin is no longer guarantees the reaching the origin of the 1 0 2 plane in a finite time,
assured. and the consequent enforcement of a second order sliding mode of the
p
This is illustrated in the following example. Consider an auxiliary controlled system.
system as in (12), (13), starting from 1 (t0 ) = 1:8, 2 (t0 ) = 2=2, Theorem 2: Consider the auxiliary system (12), (13), with the
u(t0 ) = 0 with F = 1, H = 1, G1 = 0:75, G2 = 1:25. The bounds (6)–(8) and (16), and the auxiliary control law in (20). Then,
terms which are regarded as uncertain and which respect the foregoing if (15) holds, the state of (12), (13) converges to the origin in a finite
bounds are f (t) = sin(t + 3=4), g (t) = 0:8, and, consequently, time, through the generation of a sequence of states with coordinates
h(t) = cos(t + 3=4). The controller is designed with 3 = 1, W = (M ; 0), the first component being the i-th extremal value of 1 (t)
5, according to (15), while U = 1:5 satisfies (8). Fig. 2 shows the time featuring the following contraction property:
evolution of the actual control signal after the saturation uS (t) (solid
line), and of u(t) (dashed line), while Fig. 3 shows the trajectory of the  M < j j ; i
M ; ;
= 1 2 ... (22)
Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 5, MAY 2009 1085

Case 1.2: Considering now a starting point ti with 1 (ti ) > 0 and
2 (ti ) > 0, one has that, in any case, w(ti ) = 0 3 W or w(ti ) =
0W , and this value is kept until the first switching time instant tc .
From (7), (13), (15) and (16), it follows that _ 2 (t) < 0, 8t 2 [t0 ; tc ],
that is, the trajectory of the auxiliary system moves toward the 1 -axis
an crosses it, if the control keeps its value without commutating to
w (t) = W before the 1 -axis is reached. Now, a commutation of the
control can occur in two cases:
• if 1 (t) = M , but this cannot happen, because 2 (t) > 0
implies that 1 (t) is increasing, and, due to the construction of
the controller, 1 (ti ) > M ;
• if a saturation bound is reached, but U cannot be reached because
w (t) < 0, while 2 (t) > 0, taking into account (6)–(8) and (12),
implies that 0U cannot be reached as well.
Fig. 4. Possible evolutions of the auxiliary system in Case 1.1. Note that the So, the auxiliary system state reaches the 1 -axis in a finite time
thick black horizontal line represents all the possible values of  (t ). interval, and, from this moment on, the contraction of the extremal
values will take place as described in Case 1.1. If the initial condition
is 1 (ti ) < 0 and 2 (ti ) < 0, the proof is analogous to the one
Proof: Part 1: Contraction property already seen for 1 (ti ) > 0 and 2 (ti ) > 0.
Case 1.1: Consider that the auxiliary system state, at a generic time Case 1.3: At a generic time instant, ti when 1 (ti ) < 0 and
instant ti , is in 1 (ti ) = M > 0, 2 (ti ) = 0, where M is the last 2 (ti ) > 0, the value of the auxiliary control variable can be equal

extremal value of 1 (t). It is crucial to note that at this point the system to w(ti ) = 0W or to w(ti ) = 3 W . In the first case, the auxiliary
cannot be in saturation, because if u(ti ) = U , then from (8) and (12) system evolves like in Case 1.2. In the second case, one has that
it would follow that 2 (ti ) > 0 and, for the same reason, u(ti ) = 0U M < 1 (ti ) < M , and, moreover, 1 (t) is increasing (even

would imply 2 (ti ) < 0. The generated control signal is w(ti ) = if a saturation of u(t) occurs) and is going to reach the switching
0 3 W , 2 (0; 1], and u(t) could reach its lower bound 0U before time instant after a finite time interval. After the switching occurs, the
the switching time tc . The values of  which can be obtained with behavior of the auxiliary system state is described by Case 1.2, even
the proposed control law are so that  2 [0:5; 1), and, consequently, if the state is still verifying 1 (t) < 0 and 2 (t) > 0. If 1 (ti ) > 0
2 [0:5; 1). Then, integrating (12) and (13), one obtains that, when and 2 (ti ) < 0 the proof is analogous to the one already seen for
the commutation occurs at time instant tc , such that 1 (tc ) = M , 1 (ti ) < 0 and 2 (ti ) > 0. In conclusion, starting at any point of
the value of 2 (tc ) is upper and lower bounded, as follows: the auxiliary system state, one will have 1 (t) = 0 after a finite time
interval: from this moment on, the contraction described in Case 1.1
will take place.
2 (tc ) 2 0 M ( 3 G2 W + H ); 
~2 (tc ) (23) Part 2: Finite time convergence. Considering that for any initial con-
dition the auxiliary state reaches the 1 -axis after a finite time interval,
where ~2 (tc )
 = min U 0 f (F =G1 );0 from (24) one can obtain that
2(1 0 )M ( 3 G1 W H)0 g < 0. The lower bound
is reached 0
if = 0:5, h(x(t); t) = H and g (x(t); t) = G1 at any
M <
j 01 jM j;
time instant in the considered interval, and, finally, the saturation 3 G2 0 G1 )W + 2H
value 0U is not reached. In any other case, 2 (tc ) takes a higher (

value. The upper bound  ~2 (tc ) takes into account that the absolute
 = max
2(G1 W 0 H) ; 1; 2; . . . ; j (25)

value of 2 (t) can stop increasing because of saturation of u(t), or

because a switching occurs. In both cases, since W > (H= 3 G1 ), where i , i = 1 . . . j is the value of at the corresponding iteration.
this upper bound is lower than zero. Starting from any point in interval Then, using (15) and the fact that 8i; i 2 [0:5; 1), it yields
(23), considering that 1 (tc ) = M , at tM > tc , the state
lim M = 0: (26)
trajectory crosses the abscissa axis, i.e., 2 (tM ) = 0, with j !1
The finite time convergence is not assured yet, because M could
2 0 (
3
G20 G1 )W + 2H  tend to zero only asymptotically. Let us consider a generic element
2(G1 W 0 H ) of the sequence fM g, such that 1 (tj ) = M and 2 (tj ) = 0.
1 tM M ; M : (24)

The minimum time necessary for the controller to saturate (note that
The fact that 1 (tM )  M is ensured by 2 (t)  0 8t 2 at point (1 (tj ); 2 (tj )) it is not in saturation), starting from tj ,
= 1=W , where 1 = U 0 F =G1 , while the maximum
min
[tc ; tM ], while the left extreme of the interval is obtained if the is Tsat
auxiliary state variables starts at tc with 1 (tc ) = 0:5M , 2 (tc ) = time that can elapse before the subsequent switching time instant is
0 M ( 3 G2 W + H ) (that ensures that the control variable does Tc
max
= jM j=( 3 G1 W 0 H ). The value of Tsat min
is a constant
not saturate), h(x(t); t) = H and g (x(t); t) = G2 at any time in- value which does not depend on the iteration index j , while, from
stant in the considered interval. One can verify that, imposing W > (26), one can see that limj !1 Tcmax = 0, as it occurs for the
(4H=3G1 0
3 G ), and if 2 (0; 3G =G ) holds, then j(( 3 G 0
2 1 2 2 limj !1 M , and fM g is a contractive sequence, as proved in Part
G1 )W + 2H )=(2(G1 W 0 H ))j < 1. The possible evolutions of the 1. So, the sequence fTcmax g is monotonically decreasing and goes
auxiliary system at the described iteration is shown in Fig. 4. If the ini- asymptotically to zero. For this reason, after a finite time, there will
tial condition is 1 (ti ) = M < 0, 2 (ti ) = 0 the proof is the same be an iteration ^j , when Tc
max min
< Tsat , and this will remain true for

as in the foregoing case, with reversed extremes of the intervals. This any j > ^j . After this iteration, the control signal u(t) will not saturate
ensures the contraction property. anymore. The auxiliary control law (20) generates a sequence of time
Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
1086 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 5, MAY 2009

Fig. 5. Trajectory of the auxiliary system for = 0:5. Fig. 6. Time evolution of the control signal for = 0 :5 .

instants ftM g when an extremal value of 1 (t) occurs. Each term

of this sequence is upper-bounded by the corresponding term of the
sequence

(G1 +
3 G )W
M
t^ = t^ M +
(G1 W
p 2
0 H ) 3 G2 W +H
 M : (27)

= .
From (27), recursively
Fig. 7. Trajectory of the auxiliary system for

(G1 +
3 G )W j
t^ M = p 2
0 H ) 3 G2 W
jM j + tM
(G1 W + H =1 i ensure the convergence of the proposed method. System (1) becomes
n
j x_ (t) = Ax(t) + Bu(t), where x 2 X  IR , u 2 [0U ; U ],
n2 n n
A 2 IR , B 2 IR . Define the sliding variable as a linear one,
jM j + tM
that is, 1 (t) = C T x(t), where C 2 IRm . Then, one has that the
= : (28)
i=1
uncertain terms become f (x(t); t) = C T Ax(t), h(x(t); u(t); t) =
T 2 T T
C A x(t) + C ABu(t), g (x(t); t) = C B . It is immediate to see
Then, from (25), with implicit definition of the symbols T
that, since g (x(t); t) = C B does not depend on x, condition (7)
j j is no longer needed. Bounds (6) and (16) can be found with respect to
t M <
i01 jM j + tM =
0 i01 + tM : (29)
kx(t)k, once F; H > 0 have been defined, and merged in the following
i=1 i=1 boundary condition:
F H
kC T Ak kC T A2 k
kx(t)k  max ; : (31)
Considered that assumption (15) holds and that it is trivial to see that
< 1, one can easily obtain that
Then, if the initial value of the state is sufficiently close to the origin,
0
!1 tM M
 by virtue of the stability of the sliding mode dynamics, the convergence
jlim <
10
+t (30)
is assured.
which concludes the proof. V. SIMULATION EXAMPLES
Remark 5: Note that, in this proof, it has been assumed that pa-
rameter can take any value inside the interval [0:5; ]. In practical Consider again the example of Section III, in which the origin of
applications, the designer can make different decisions: for example, the 1 0 2 -plane was not reached using the conventional algorithm
one can set = 0:5 for all iterations, which means to adopt a strategy (see, Figs. 2 and 3), and apply the control strategy (20), using, at any
consisting of the conventional sub-optimal auxiliary control law plus iteration, first = 0:5, and then = .
the desaturation strategy, to cope with the presence of saturating ac- Using = 0:5, the auxiliary system state converges to the origin
tuators. In alternative, one can set = , that is, a switching of the in a finite time as shown in Fig. 5, while the time evolution of u(t)
auxiliary control law occurs as soon as a saturation bound is reached. is shown in Fig. 6. Using the choice of = , the controller never
These two solutions (which represent particular cases of the generic produces actuator saturation, but, during the reaching phase, its output
algorithm) and their effects are analyzed in the simulation examples u(t) oscillates between one of its maximum values 6U , and the min-

reported in the next section. Any other admissible choice of the value imum value that must be used to counteract the uncertain term, that is,
of would produce effects which would be intermediate between the 0f (x(t); t)=g (x(t); t). The convergence of the auxiliary system state
described ones. In particular, when is near to 0.5 the control variable is shown in Fig. 7, while the time evolution of u(t) is shown in Fig. 8.
u(t) can stay in saturation for rather long time intervals, and this could
cause problems or damages in some classes of actuators (for example, VI. CONCLUSION
electrical machines are very sensitive to current saturations), but the The technical note addresses the problem of the possible saturation
control action usually leads to short reaching time. On the other hand, of the continuous control variable which can occur when a the sub-op-
if is near to , the length of the time intervals when u(t) is saturated timal SOSM controller is used to control systems, with which, because
is reduced, but the reaching times can be quite long, if the initial con- of the choice of the sliding variable, a relative degree one auxiliary
dition 1 (t0 ) is very far from the origin. system can be associated. The reaching phase and the sliding phase are
Remark 6: If the considered system is a linear one with saturating analyzed, showing that actuator saturation can occur only during the
actuators, a boundary condition on kx(t)k can be found, in order to first one. Nevertheless, when saturation occurs, problems can arise as
Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 5, MAY 2009 1087

A Parametrized Controller Reduction Technique via a New

Frequency Weighted Model Reduction Formulation
Pantazis Houlis and Victor Sreeram

Abstract—In this technical note, the solution to the controller reduction

problem via a double-sided frequency weighted model reduction technique
is considered. A new method for finding low-order controllers based on new
Fig. 8. Time evolution of the control signal for = . frequency weights derived using closed-loop system approximation crite-
rion is proposed. The formulas of frequency weights are obtained in terms
of the plant, the original controller and a matrix of free parameters. By
varying the free parameters in the resulting two-sided frequency weighted
far as the convergence to zero of the sliding variable is concerned, pre- model reduction problem, frequency weighted error can be significantly re-
venting the enforcement of the desired second order sliding mode. In duced to yield more accurate low-order controllers.
this technical note, a modified sub-optimal SOSM control algorithm is Index Terms—Controller reduction, frequency weights, model reduction.
proposed. The modification is oriented to avoid the delay in the con-
troller switching caused by actuator saturation. The proposed controller
proves to guarantee the convergence of the sliding variable and of its I. INTRODUCTION
first time derivative to zero in a finite time, in spite of the presence of Controller reduction problems are usually solved via frequency
uncertain terms affecting the system model and of the saturating actu- weighted model reduction problem [1]–[6]. The frequency weighted
ators. model reduction problem can be classified into single-sided or
double-sided frequency weighted problems. The single-sided fre-
REFERENCES quency weighted model reduction problem is based on stability
[1] T. Hu and Z. Lin, Control Systems With Actuator Saturation: Analysis margin considerations. The reduced-order controller should satisfy the
and Design. Boston, MA: Birkhauser, 2001.
same conditions as are listed in the above references. The double-sided
[2] K. S. Åström and T. Hagglund, PID Controllers: Theory, Design and
Tuning. Research Triangle Park, NC: Instrument Society of America, problem is based on closed-loop system approximation and attempts
1995. to minimize an index of the form
[3] D. Q. Mayne, J. B. Rawlings, C. V. Rao, and P. O. M. Scokaert,
“Constrained model predictive control: Stability and optimality,”
Automatica, vol. 36, no. 6, pp. 789–814, 2000. = kV1 (K 0 Kr )V2 k1
e
where V1 = (I + GK )01 G and = (I + GK )01
[4] V. I. Utkin, Sliding Mode in Control and Optimization. Berlin, Ger-
many: Springer-Verlag, 1992. V2
[5] V. I. Utkin, J. Guldner, and J. Shi, Sliding Mode Control in Electro-
mechanical Systems. London, U.K.: Taylor & Francis, 1999.
[6] C. Edwards and K. S. Spurgeon, Sliding Mode Control: Theory and and G, K , and Kr are the plant, the original controller and the reduced
Applications. London, U.K.: Taylor & Francis, 1998. controller respectively.
[7] V. I. Utkin and H. Lee, “Chattering problem in sliding mode control There are a few methods for the solution of the frequency weighted
systems,” in Proc. Int. Workshop Variable Struct. Syst., Alghero, Italy, model reduction problem [4], [5]. However, approximation errors ob-
Jun. 2006, pp. 346–350.
[8] I. Boiko, “On fractal nature of parasitic dynamics in sliding mode tained using these techniques are large. In general, the techniques are
control,” in Proc. Amer. Control Conf., New York, Jul. 2007, pp. not as good as the techniques available for the unweighted case [7].
3819–3824. In this technical note, new techniques for obtaining lower approxi-
[9] A. Levant, “Chattering analysis,” in Proc. Eur. Control Conf., Kos, mation error using standard techniques by manipulating the frequency
Greece, Jul. 2007, pp. 3195–3202. weights are proposed [4], [5], [8]–[12]. The techniques are based on de-
[10] I. Boiko, L. Fridman, A. Pisano, and E. Usai, “Performance analysis of
second-order sliding-mode control systems with fast actuators,” IEEE riving a new set of weights for double-sided frequency weighted model
Trans. Automat. Control, vol. 52, no. 6, pp. 1053–1059, Jun. 2007. reduction. Out of the two new weights derived, one of the weights can
[11] A. Levant, “Sliding order and sliding accuracy in sliding mode control,” be made to be a function of free parameters. By varying those free
Int. J. Control, vol. 58, no. 6, pp. 1247–1263, 1993. parameters in the resulting double-sided frequency weighted model
[12] L. Fridman and A. Levant, “Higher order sliding modes,” in Sliding
Mode Control in Engineering, ser. Control Engineering Series, W. Per- reduction problem, the frequency weighted error can be significantly
ruquetti and J. P. Barbot, Eds. London, U.K.: CRC Press, 2002. reduced, subsequently obtaining more accurate low-order controllers.
[13] G. Bartolini, A. Ferrara, F. Levant, and E. Usai, “On second order Note that this technical note contains an improved and generalized ver-
sliding mode controllers,” in Variable Structure Systems, Sliding Mode sion of the method presented in [13], [14].
and Nonlinear Control, ser. Lecture Notes in Control and Informa- Preliminary results on the Double Controller Techniques that were
tion Sciences, K. D. Young and U. Özgüner, Eds. Berlin, Germany:
Springer-Verlag, 1999. presented in [13], were only applicable to SISO (Single Input Single
[14] G. Bartolini, A. Ferrara, and E. Usai, “Output tracking control of un- Output) systems. The method presented here is a generalization which
certain nonlinear second-order systems,” Automatica, vol. 33, no. 12,
pp. 2203–2212, 1997.
[15] G. Bartolini, A. Ferrara, and E. Usai, “Chattering avoidance by second-
order sliding mode control,” IEEE Trans. Automat. Control, vol. 43, no. Manuscript received September 07, 2005; revised October 09, 2006. Current
2, pp. 241–246, Feb. 1998. version published May 13, 2009. Recommended by Associate Editor C. Beck.
[16] G. Bartolini, A. Ferrara, E. Usai, and V. Utkin, “On multi-input chat- The authors are with the School of Electrical and Electronic Engineering,Uni-
tering-free second-order sliding mode control,” IEEE Trans. Automat. versity of Western Australia, Crawley WA 6009, Australia (e-mail: sreeram@ee.
Control, vol. 45, no. 9, pp. 1711–1717, Sep. 2000. uwa.edu.au).
[17] A. Ferrara and M. Rubagotti, “A sub-optimal second order sliding Color versions of one or more of the figures in this technical note are available
mode controller for systems with saturating actuators,” in Proc. Amer. online at http://ieeexplore.ieee.org.
Control Conf., Seattle, WA, Jun. 2008, pp. 4715–4720. Digital Object Identifier 10.1109/TAC.2008.2010993

0018-9286/$25.00 © 2008 IEEE

Authorized licensed use limited to: Nazarbayev University. Downloaded on February 18,2025 at 09:34:11 UTC from IEEE Xplore. Restrictions apply.