Automatica 109 (2019) 108515

Contents lists available at ScienceDirect

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

Brief paper

Continuous nonsingular terminal sliding mode control based on

adaptive sliding mode disturbance observer for uncertain nonlinear
systems✩
∗
Hamed Rabiee, Mohammad Ataei , Mohsen Ekramian
Department of Electrical Engineering, University of Isfahan, Isfahan, Iran

article info a b s t r a c t

Article history: In this paper, a finite time control method for an uncertain nonlinear system is proposed. An
Received 13 March 2018 adaptive sliding mode disturbance observer is designed to estimate the disturbance in finite time.
Received in revised form 19 February 2019 The assumptions on the disturbance are relaxed in the sense that, its first derivative upper bound
Accepted 7 July 2019
is considered to be unknown and only the order of its second derivative upper bound is known.
Available online xxxx
Based on the output of proposed disturbance observer, a terminal sliding mode control scheme for
Keywords: the uncertain nonlinear system is presented. The designed control law is continuous and nonsingular.
Sliding mode disturbance observer Furthermore, compared with some existing sliding mode controllers, the conditions on the controller
Nonsingular terminal sliding mode parameters bounds are relaxed. It is proved that the disturbance observer error as well as the system
Continuous control states converges to the origin in finite time. Finally, the effectiveness of the proposed method is shown
Uncertain nonlinear systems
by numerical simulations.
© 2019 Elsevier Ltd. All rights reserved.

1. Introduction Although TSMC is faster than conventional SMC, it usually

encounters three main issues. First, the control law is not con-
In recent decades, many researchers have paid attention to the tinuous and therefore causes chattering. Second, the control law
sliding mode control(SMC) because of its robustness to distur- has a singular point and third, it is assumed that the upper bounds
bances and uncertainties, finite time convergence and design sim- on the uncertainty and disturbance are known. In Feng, Yu, and
plicity. This method is applied to control many systems including Man (2002), a nonsingular terminal sliding mode control(NTSMC)
industrial robot manipulators, underwater vehicles, spacecraft, is proposed by changing the sliding surface, but the control law
switching power supplies and electrical motors. The SMC involves is still not continuous. In Yu, Yu, Shirinzadeh, and Man (2005),
two phases namely the reaching and sliding phases respectively. continuous NTSMC is presented, however the convergence of
In the reaching phase, the controller operates so that the system the states in finite time is only guarantied where there is no
states reach the sliding surface in finite time. Then, in the sliding uncertainty and disturbance. In Feng, Han, and Yu (2014, 2016),
phase the states maintain on the sliding surface and consequently a terminal sliding mode controller is proposed that its control
the states converge to the origin. In conventional SMC the sliding law is continuous and nonsingular, but the upper bounds of
surface is designed linear, therefore, the sliding motion to the disturbance and its first derivative are assumed to be known.
origin is accomplished asymptotically. In recent years, terminal Recently, the disturbance observer-based methods is utilized
sliding mode control(TSMC) (Zhihong, Paplinski, & Wu, 1994; to counteract disturbances (Chen, Yang, Guo, & Li, 2016; Li, Yang,
Zhihong & Yu, 1997) has been introduced in which the sliding Chen, & Chen, 2014). In these methods, at first, an estimation of
surface is designed nonlinear and makes the finite time state the disturbance is calculated by a disturbance observer, then it is
convergence. used in the control law. Moreover the disturbance observer can
be used to reduce the chattering in sliding mode control.
In Chen and Chen (2010), Yang, Li, and Yu (2013), a distur-
✩ The material in this paper was not presented at any conference. This paper
bance observer-based conventional SMC is presented. However,
was recommended for publication in revised form by Associate Editor Hyungbo
in Chen and Chen (2010) it is assumed that the derivative of the
Shim under the direction of Editor Daniel Liberzon.
∗ Corresponding author. disturbance is zero, and in Yang et al. (2013) it is assumed that
E-mail addresses: h.rabiee@eng.ui.ac.ir (H. Rabiee), ataei@eng.ui.ac.ir the derivative of disturbance satisfies limt →∞ ḋ = 0. In Su, Yang,
(M. Ataei), m.ekramian@eng.ui.ac.ir (M. Ekramian). and Li (2014), a disturbance observer and the continuous NTSMC

https://doi.org/10.1016/j.automatica.2019.108515
0005-1098/© 2019 Elsevier Ltd. All rights reserved.
2 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515

in Yu et al. (2005) are combined where the bounds on disturbance Theorem 3 (Edwards & Shtessel, 2016). Consider the prototypical
derivative are assumed to be known. In Ginoya, Shendge, and dynamic system with a first order sliding mode controller. For this
Phadke (2014), Li, Sun, Yang, and Yu (2015), Yang, Li, Su and system the following dynamic of sliding surface exists:
Yu (2013), Yang, Su, Li and Yu (2014), disturbance observer is
used to estimate the matched and mismatched disturbances, but σ̇a = a(t) + ua (t) (4)
it is also assumed that the upper bounds of the disturbance and
where a(t) represents the uncertainty and satisfies the conditions
its derivatives are known. However, the upper bounds of the
|a(t)| < a0 and |ȧ(t)| < a1 , where the bounds a0 and a1 are
disturbance and its derivatives are usually unknown in practical
unknown constants. Consider the following control input :
systems.
The above mentioned issue motivated us to propose an adap- ua (t) = −(ka (t) + ηa )sign(σa ) (5)
tive disturbance observer-based continuous NTSMC. Compared
with (Ginoya et al., 2014; Li et al., 2015; Su et al., 2014; Yang, Li where ηa is a small positive constant and ka (t) is updated by the
et al., 2013; Yang, Su et al., 2014), the proposed adaptive sliding following two layers adaptive law :
mode disturbance observer(ASMDO) does not require the upper
bound of disturbance derivative and only the order of second k̇a (t) = −ρa (t)sign(δa (t)) (6)
γa |δa (t)| if |δa (t)| > δ0a
{
derivative upper bound is known. Also, compared with some
ṙa (t) = (7)
existing sliding mode controllers as in Feng et al. (2014), the 0 otherwise
assumption on disturbance in the proposed ASMDO is less strict,
because it does not consider any bound on the disturbance and where
therefore it includes a wider range of disturbances(for example ρa (t) = r0a + ra (t) (8)
in our proposed ASMDO the disturbance can be unbounded).
The designed control law is continuous which avoids chattering 1
δa (t) = ka (t) − |ūeq−a | − ϵa (9)
and is also nonsingular. Furthermore, this control law ensures αa
that the states converge to origin in finite time. These features 1
ū˙ eq−a = (ua (t) − ūeq−a ) (10)
are achieved while the bounds on the controller parameters as τa
in Feng et al. (2014) are relaxed.
The structure of this paper is as follows. In Section 2, the prob- If the positive constants 0 < ) αa < 1, r0a , δ0a , γa , ϵa , τa and
qa > sup 1, | dtd (ūeq−a (t))|/a1 are chosen such that the inequality
(
lem and required mathematical preliminaries are presented. In
Section 3, the ASMDO is expressed. Then continuous nonsingular ( )2
1 1 qa a1
terminal sliding mode control(CNTSMC) scheme is introduced in ϵa2 > δ0a
2
+ (11)
Section 4, and in Section 5, the proposed controller and distur- 4 γa αa
bance observer are simulated on an uncertain nonlinear system. holds, then the condition ka (t) > |a(t)| is achieved in finite time,
Finally, the concluding remarks are presented in Section 6. which guarantees a persistent sliding motion. Furthermore, the gains
ka (t) and ρa (t) are bounded.
2. Problem statement and mathematical preliminaries
Remark 4. In order to satisfy the inequality (11), the bound a1 is
Consider the following uncertain nonlinear system:
⎧ not required to be known exactly. In fact, for given δ0a , αa and qa ,
⎪ ẋ1 = x2 just by knowing the order of a1 , the constant γa can be chosen
⎨ ẋ2 = x3 sufficiently large so there exists an ϵa for which the inequality
⎪
⎪
⎪
.. (1) (11) holds (Edwards & Shtessel, 2016).
⎪ .
ẋ = xn
⎪
n − 1
⎪
Remark 5 (Edwards & Shtessel, 2016). The equivalent control
⎪
ẋn = f (x) + g(x)u + d(x, t)
⎩
input ueq−a for sliding surface σa is a(t). Based on the adaptive
where x = [x1 , . . . , xn ]T is the state vector, f (x) and g(x) ̸ = 0 : ∀x law presented in Eqs. (6)–(10), ka is increased until sliding mo-
are smooth nonlinear functions and d(x, t) is the total disturbance tion occurs, then ka starts to decrease so that it converges to a
including the external disturbance and uncertainty effects. The safety margin. This margin is around ueq−a and depends on the
following assumption is considered for the disturbance d(x, t). parameters ϵa and αa , as follows.

Assumption 1. The first and second derivatives of the distur- 1 1

ka (t) ≈ |ueq−a | + ϵa = |a(t)| + ϵa (12)
bance exist as |ḋ| < L1 , |d̈| < L2 , where the upper bound L1 is αa αa
assumed to be unknown and only the order of L2 is known.
Lemma 6 (Bhat & Bernstein, 2005). Consider the following sliding
The purpose is to design an ASMDO with a terminal sliding
surface for system (1):
mode controller, to estimate the disturbance d and converge the
(n) (n−1) (n−1) αn
Routh-locus Hurwitz
α1
system states (1) to the origin both in finite time. s = x1 + cn sgn(x1 )|x1 | + · · · + c1 sgn(xCriteria
1 )|x1 |
The following theorem and lemmas are used in the rest of
article. = ẋn + cn sgn(xn )|xn |αn + · · · + c1 sgn(x1 )|x1 |α1 (13)

where αi and ci i = 1, 2, . . . , n are some constants that are chosen

Lemma 2 (Zuo & Tie, 2016). Consider the following scalar system:
so that the polynomial pn + cn pn−1 + · · · + c2 p + c1 is Hurwitz,
ẏ = −c1 sgn(y)|y|α1 − c2 sgn(y)|y|α2 y(0) = y0 (2) i.e. all of its roots are in the left-half plane. Also, αi s are determined
by following equations:
where c1 , c2 are positive constants, 0 < α1 < 1 and α2 > 1. The
α1 = α, n = 1
{
equilibrium point of the above system is finite time stable and its
αα 1 (14)
settling time is bounded by : αi−1 = 2α i i+−α , i = 2, . . . , n ∀n ≥ 2
i+1 i
1 1 1 1 where αn+1 = 1, αn = α , α ∈ (1 − ϵ, 1), ϵ ∈ (0, 1). If u is
T < Tmax ≜ + (3)
c1 1 − α1 c2 α2 − 1 determined in such a way that the system (1) reaches the sliding
 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515 3

surface, i.e. s = 0, then the dynamic of the system (1) becomes: Now consider the Lyapunov function candidate V = σd2 /2. By
⎧ differentiating V and substituting (23), one achieves:
⎪ ẋ1 = x2
⎪ ẋ2 = x3 V̇ = σd σ̇d = −ηd |σd | − kd |σd | − ḋσd
⎪
⎪
..
⎨
. (15) ⩽ (|ḋ| − kd )|σd | − ηd |σd | (24)
⎪
⎪ ẋn−1 = xn
⎪
Since ηd > 0 and kd > L1 , it is concluded that:
⎪
ẋn = −cn sgn(xn )|xn |αn − · · · − c1 sgn(x1 )|x1 |α1
⎩

and the states of the system (15) converge to the zero equilibrium σd σ̇d ⩽ −ηd |σd |
point x = 0 from any arbitrary initial point x(0) ̸ = 0 in finite time. which yields the convergence of sliding surface σd to zero in finite
time i.e. σd = 0. As a result, by using Eq. (18) one gets:
Definition 7 (Li et al., 2014). Consider a nonlinear system ẋ =
f(t , x, w) where f : [0, ∞]× Rn × Rm → Rn is piecewise continuous ė = −c1d sgn(e)|e|α1d − c2d sgn(e)|e|α2d (25)
in t and locally Lipschitz in x and w. This system is said to be
By considering Lemma 2, the equilibrium of (25) as e = 0 is finite
locally input-to-state stable(ISS), if there exist a class KL function
time stable.
χ , a class K function κ , and positive constants k1 and k2 such that
Now consider the estimation error of disturbance d as follows:
for any initial state x(t0 ) with ∥x(t0 )∥ < k1 and any input w(t)
with supt ⩾t0 ∥w(t)∥ < k2 , the solution x(t) exists and satisfies : ed = d − d̂ (26)
∥x(t)∥ ⩽ χ (∥x(t0 )∥, t − t0 ) + κ ( sup ∥w(t)∥) Finally by using (22) and (20), it is concluded that ed = 0. As a
t0 ⩽τ ⩽t
result, the proposed observer (16)–(20) estimates the disturbance
for all t ⩾ t0 ⩾ 0. d in finite time. ■

Lemma 8 (Li et al., 2014). Consider the system ẋ = F(x, w) = In Theorem 9, it is shown that if kd > L1 , then the proposed
f(x) + p(x)w where F : D × Dw → Rn is locally Lipschitz in x, disturbance observer estimates the disturbance d in finite time.
D ∈ Rn is a domain that contains x = 0, and Dw ∈ Rm is a domain However, in practical problems, the exact upper bound L1 is
that contains w = 0. If p(x) is continuously differentiable and the unknown. As proposition 2 in Edwards and Shtessel (2016), in the
origin of system ẋ = f(x) is asymptotically stable, then ẋ = F(x, w) next theorem, the observer is improved in an adaptive scheme to
is locally ISS. overcome this drawback.

3. Adaptive sliding mode disturbance observer (ASMDO) Theorem 10. Consider the disturbance observer in Theorem 9 as
(16)–(20) in which the gain kd is updated by the following two layers
adaptive law :
In this section, in order to design an ASMDO for system (1),
the following auxiliary variable is introduced: k̇d (t) = −ρd (t)sign(δd (t)) (27)
γd |δd (t)| if |δd (t)| > δ0d
{
e = z − xn (16)
ṙd (t) = (28)
0 otherwise
where the variable z has the following dynamic:
where
ż = f (x) + g(x)u − c1d sgn(e)|e|α1d − c2d sgn(e)|e|α2d + vz (17)
ρd (t) = r0d + rd (t) (29)
The coefficients c1d , c2d , α1d and α2d are positive constants with
1
α1d < 1 and α2d > 1. In the following theorem, vz is designed δd (t) = kd (t) − |ūeq−d | − ϵd (30)
such that the auxiliary variable e in (16) as well as estimation αd
error of disturbance d converges to zero in finite time. 1
ū˙ eq−d = (−(ηd + kd )sgn(σd ) − ūeq−d ) (31)
τd
Theorem 9. Consider the following sliding surface σd :
If the constants 0 < αd < 1, δ0d , γd , ϵd and qd > sup(1, | dtd (ūeq−d
α1d α2d
σd = ė + c1d sgn(e)|e| + c2d sgn(e)|e| (18) (t))|/L2 ) are chosen such that the inequality
)2
If vz is designed by the following dynamic equation:
(
1 1 qd L2
ϵd2 > δ0d
2
+ (32)
v̇z = −(ηd + kd )sgn(σd ), vz (0) = 0 (19) 4 γd αd
holds, then the condition kd (t) > |L1 | is achieved in finite time, which
where ηd is a positive constant and kd satisfies the condition kd > L1 ,
guaranties a persistent sliding motion. Furthermore, the gains kd (t)
then e converges to zero in finite time and the finite time estimation
and ρd (t) are bounded.
of the disturbance d is obtained as follows:

d̂ = vz (20) Proof. Considering Assumption 1, the first and second derivatives

of the disturbance d(t) are bounded. By following Theorem 3 and
Proof. First, consider the derivative of e as follows: replacing a(t) and σa in (4) by −ḋ and σd in (23) respectively, the
two layers adaptive law in (27)–(31) can be employed to achieve
ė = ż − ẋn = −c1d sgn(e)|e|α1d − c2d sgn(e)|e|α2d + vz − d (21) the results. ■
By substituting Eq. (21) into (18) we obtain:
Remark 11. Considering remark 2.3 in Feng et al. (2014), in
σd = v z − d (22) the proposed observer in Theorems 9 and 10, σd in (18) cannot
be obtained because ẋn is not available. However, in order to
Using (19), the derivative of the sliding surface σd is given by:
estimate the disturbance, sgn(σd ) in (19) is required which can
σ̇d = v̇z − ḋ = −(ηd + kd )sgn(σd ) − ḋ (23) be achieved as follows. By integrating both sides of (18), we
4 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515

have In the next step, it is shown that |un (t)| < |ed | ∀t. Accordingly,
∫ t by using Eq. (37) we have un (0) ∈ [−|ed |, |ed |]. In the case of
g(t) = σd (τ )dτ un > |ed |, Eq. (39) yields:
0
s = un + d − d̂ = un + ed > 0
∫ t
=e+ (c1d sgn(e)|e|α1d + c2d sgn(e)|e|α2d ) dτ (33)
0 Moreover, by using Eqs. (37) and (38) , it can be concluded that:
which yields σd (t) = limτ →0 (g(t) − g(t − τ )) /τ . Thus, sgn(σd )
u̇n = −Tun + us = −Tun − (η + T |un |)sgn(s) < 0
can be obtained as follows:
In fact, for un > |ed |, u̇n un < 0. Similarly, for the case un < −|ed |,
sgn(σd ) = sgn (g(t) − g(t − τ )) (34)
one achieves s < 0 and u̇n > 0 which yields u̇n un < 0. Thus,
where τ is a short time delay which can be considered as the one can conclude that un ∈ [−|ed |, |ed |] ∀t, i.e. un is bounded.
fundamental sample time. Moreover, one can conclude that:

Remark 12. Similar to Section 3, inequality (32) is only a |un + d − d̂| < |un | + |d − d̂| < 2|ed | (41)
sufficient condition and in order to satisfy it, the bound L2 is not Now by applying the control law (35) and (36) into system (1),
required to be known exactly and by selecting the adaptive gain the closed loop system dynamic is given by:
γd sufficiently large (to dominate L2 ), the inequality (32) can be
satisfied (see Remark 4). ẋ = f̄(x) + p̄ū (42)

where f̄ and p̄ are:

4. Continuous nonsingular terminal sliding mode control
based on adaptive sliding mode disturbance observer x2
⎡ ⎤
⎢ x3 ⎥
In this section by using the proposed disturbance observer in
⎢
f̄ = ⎢
.. ⎥
.
⎥
Theorem 10 and the sliding surface stated in Lemma 6, a CNTSMC ⎢ ⎥
xn
⎣ ⎦
for system (1) is proposed.
−cn sgn(xn )|xn |αn − · · · − c1 sgn(x1 )|x1 |α1
Theorem 13. Consider the nonlinear system (1) and the sliding
p̄T = 0 0 · · · 0 1
[ ]
surface s in (13). The states of system converge to the origin in finite
time if the control law u is designed as and the input ū = un + d − d̂, as in (41), is bounded. Note
u=g −1
(x)(ueq + un ) (35) that, based on Lemma 6 the system ẋ = f̄(x) is finite-time stable
and therefore it is asymptotically stable. Finally, the system (42)
where satisfies the conditions of Lemma 8, thus it is locally ISS. This
ensures the state x is bounded in t < Tconv . ■
ueq = −f (x) − cn sgn(xn )|xn |αn − · · · − c1 sgn(x1 )|x1 |α1 − d̂ (36)

and d̂ is estimated by the disturbance observer in Theorem 10. Also Remark 14. Similar to the proposed observer in Theorems 9 and
10, the sliding surface s in the proposed controller in Theorem 13
u̇n + Tun = us , un (0) = 0 (37) is not directly available (see Remark 11). In order to construct
sgn(s) in the control law (38) one can employ

us = −(η + T |un |)sgn(s) (38) sgn(s) = sgn (h(t) − h(t − τ ))

where T and η are some positive constants. where

h(t) = xn +
Proof. By substituting Eqs. (35), (36) and ẋn into (13), the sliding ∫ t
surface s can be written as: (cn sgn(xn )|xn |αn + · · · + c1 sgn(x1 )|x1 |α1 )dτ
0
s = un + d − d̂ (39)
Assume that the proposed disturbance observer converges in Remark 15. Considering equations (35)–(38), the proposed con-
t = Tconv , then for t ⩾ Tconv the sliding surface is s = un . By trol law is continuous and nonsingular. Furthermore, according
differentiating the sliding surface and substituting Eqs. (37) and to |un (t)| < |ed | ∀t, it can be concluded that for t > Tconv
(38), the following equation is obtained: (the convergence time of disturbance observer), we have us =
−η sgn(s) in (38) where the switching coefficient η can be se-
ṡ = us − Tun = −(η + T |un |)sgn(s) − Tun (40)
lected as a small value. In fact, compared with some existing
Now consider the Lyapunov function candidate V = s /2. By 2 sliding mode controllers as in Feng et al. (2014), the conditions
differentiating V and substituting (40), one achieves : on the controller parameters bounds are relaxed.

V̇ = sṡ = −η|s| − T |un ||s| − Tun s

5. Numerical simulation
⩽ −η|s| − T |un ||s| + T |un ||s| ⩽ −η|s|

which yields the convergence of sliding surface s to zero (s = 0) Consider the dynamic model of Van der Pol circuit as fol-
in finite time. Moreover, according to Lemma 6, the state x will lows (Chen, Wu, & Cui, 2013):
converge to the origin in finite time. {
ẋ1 = x2
In the following, it is proved that for t < Tconv , the states re-
ẋ2 = −2x1 + 3(1 − x21 )x2 + u(t) + d(t)
main bounded. First, note that the proposed disturbance observer
in Theorem 10 converges in finite time which implies that the This system is simulated with two types of disturbances, type 1:
error ed in (26) is bounded. d1 (t) = 2 sin(0.2π t) + 0.15 sin(2π t) and type 2: d2 (t) = u(t)
 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515 5

Fig. 1. Disturbance d1 (t) and its estimation. Fig. 4. Evolutions of kd (t) and |ḋ1 (t)|.

Fig. 2. Estimation error of disturbance d1 (t).

Fig. 5. Control input(for d1 (t)).

Fig. 3. Evolution of ρd (for d1 (t)).

Fig. 6. Evolutions of system states(for d1 (t)).

(step). In order to apply the proposed method, the sliding surface

σd and the parameters of the ASMDO are selected as: The initial conditions for system and disturbance observer are
0.5 x1 (0) = 1, x2 (0) = −1, and z(0) = 0 respectively.
σd = ė + 2sgn(e)|e| + 2sgn(e)|e| 3
The simulation results are presented in Figs. 1–6 for d1 (t)
ηd = 0.01, τd = 0.001, ϵd = 0.65 and 7–12 for d2 (t). Note that, the estimated disturbances d̂1 (t)
αd = 0.999, δ0d = 0.09, γd = 470, r0d = 0.1 in 1 and d̂2 (t) in 7 coincide with disturbances d1 (t) and d2 (t)
respectively, while the disturbance errors ed1 (t) in 2 and ed2 (t)
Moreover the controller parameters and sliding surface s(t) are in 8 converge to zero in finite time. The bounded gain ρd (t) is
given by: shown in Figs. 3 and 9. Moreover, in Figs. 4 and 10, the adaptive
7 7 gain kd (t) is shown which varies relative to |ḋ1 (t)| and |ḋ2 (t)|. As
s = ẋ2 + 14sgn(x2 )|x2 | 10 + 49sgn(x1 )|x1 | 13
in Remark 5, kd (t) converges to a safety margin which depends
T = 0.1, η = 0.001 on the parameters ϵd and αd . Finally, by applying the continuous
6 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515

Fig. 7. Disturbance d2 (t) and its estimation. Fig. 10. Evolutions of kd (t) and |ḋ2 (t)|.

Fig. 8. Estimation error of disturbance d2 (t).

Fig. 11. Control input(for d2 (t)).

Fig. 9. Evolution of ρd (for d2 (t)).

Fig. 12. Evolutions of system states(for d2 (t)).

nonsingular control law u(t) in Figs. 5 and 11, the system states
in Figs. 6 and 12 converge to the origin in finite time. 6. Conclusion
Note that the step type disturbance is not in the group of
disturbances discussed in the our problem, since this type of In this paper, a disturbance observer-based control method
disturbance does not satisfy the condition |ḋ(t)| < L1 (the first for an uncertain nonlinear system is presented. First, an adaptive
derivative of step does not exist). However, as it is clear in sliding mode disturbance observer is proposed which estimates
Fig. 7–12 the proposed method can successfully estimate the step the disturbance in finite time. In observer synthesis method,
type disturbance and converge the system to zero in finite time. the upper bounds of the first and second derivatives of the
Moreover, note that the existing step type disturbance in practice disturbance are considered to be unknown. In order to stabi-
usually does not involve jump and due to continuity nature, one lize the closed loop uncertain system, the finite time control
can apply the proposed method in this case by considering the method based on the proposed disturbance observer is then
sufficiently large upper bound for |ḋ| proposed. This control law is continuous and nonsingular that
 H. Rabiee, M. Ataei and M. Ekramian / Automatica 109 (2019) 108515 7

causes the system states converge to origin in finite time. Further- Yu, S., Yu, X., Shirinzadeh, B., & Man, Z. (2005). Continuous finite-time control
more, compared with some existing sliding mode controllers, the for robotic manipulators with terminal sliding mode. Automatica, 41(11),
1957–1964.
conditions on controller parameters bounds are relaxed. Finally,
Zhihong, M., Paplinski, A. P., & Wu, H. R. (1994). A robust mimo terminal sliding
the numerical simulations approve the advantages of proposed mode control scheme for rigid robotic manipulators. IEEE Transactions on
method. Automatic Control, 39(12), 2464–2469.
Zhihong, M., & Yu, X. H. (1997). Terminal sliding mode control of mimo linear
References systems. IEEE Transactions on Circuits and Systems I, 44(11), 1065–1070.
Zuo, Z., & Tie, L. (2016). Distributed robust finite-time nonlinear consensus
protocols for multi-agent systems. International Journal of Systems Science,
Bhat, S. P., & Bernstein, D. S. (2005). Geometric homogeneity with applications
47(6), 1366–1375.
to finite-time stability. Mathematics of Control, Signals, and Systems, 17(2),
101–127.
Chen, M., & Chen, W.-H. (2010). Sliding mode control for a class of uncertain
nonlinear system based on disturbance observer. International Journal of Hamed Rabiee received B.S. in Electronics Engineering
Adaptive Control and Signal Processing, 24(1), 51–64. from Malek-Ashtar University of Technology, Isfahan,
Chen, M., Wu, Q.-X., & Cui, R.-X. (2013). Terminal sliding mode tracking control Iran, in 2008 and the M.S. in Control Engineering from
for a class of siso uncertain nonlinear systems. ISA Transactions, 52(2), Isfahan University of Technology, Isfahan, Iran, in 2010.
198–206. He is currently a Ph.D. candidate in Control Engineer-
Chen, W. H., Yang, J., Guo, L., & Li, S. (2016). Disturbance-observer-based control ing, at the University of Isfahan. His research interests
and related methods; an overview. IEEE Transactions on Industrial Electronics, include sliding mode control/observation, disturbance
63(2), 1083–1095. observer and control of robot manipulator.
Edwards, C., & Shtessel, Y. B. (2016). Adaptive continuous higher order sliding
mode control. Automatica, 65, 183–190.
Feng, Y., Han, F., & Yu, X. (2014). Chattering free full-order sliding-mode control.
Automatica, 50(4), 1310–1314.
Feng, Y., Han, F., & Yu, X. (2016). Reply to comments on ‘chattering free full-
Mohammad Ataei received the B.S. degree from the
order sliding-mode control’ [automatica 50 (2014) 1310–1314]. Automatica,
Isfahan University of Technology, Iran, in 1994, the
72, 255–256.
M.S. degree from the Iran University of Science &
Feng, Y., Yu, X., & Man, Z. (2002). Non-singular terminal sliding mode control
Technology, Iran, in 1997, and Ph.D. degree from K.
of rigid manipulators. Automatica, 38(12), 2159–2167.
N. Toosi University of Technology, Iran, in 2004 (joint
Ginoya, D., Shendge, P. D., & Phadke, S. B. (2014). Sliding mode control for
project with the University of Bremen, Germany under
mismatched uncertain systems using an extended disturbance observer. IEEE
DAAD scholarship) all in control Engineering. Since
Transactions on Industrial Electronics, 61(4), 1983–1992.
2004, he has been with the Department of Electrical
Li, S., Sun, H., Yang, J., & Yu, X. (2015). Continuous finite-time output regulation
Engineering at the University of Isfahan, Iran, where he
for disturbed systems under mismatching condition. IEEE Transactions on
is currently serving as professor. His main areas of re-
Automatic Control, 60(1), 277–282.
search interest include control theory and applications,
Li, S., Yang, J., Chen, W.-H., & Chen, X. (2014). Disturbance Observer-Based Control:
nonlinear control and observers, chaotic systems’ analysis and control.
Methods and Applications (1st ed.). Boca Raton, FL, USA: CRC Press, Inc.
Su, J., Yang, J., & Li, S. (2014). Continuous finite-time anti-disturbance control
for a class of uncertain nonlinear systems. Transactions of the Institute of
Measurement and Control , 36(3), 300–311. Mohsen Ekramian was born in 1984 in Isfahan, Iran.
Yang, J., Li, S., Su, J., & Yu, X. (2013). Continuous nonsingular terminal sliding He received his B.Sc., M.Sc. and Ph.D. degrees in Elec-
mode control for systems with mismatched disturbances. Automatica, 49(7), trical (Control) Engineering from Isfahan University of
2287–2291. Technology, in 2005, 2007 and 2011, respectively. Since
Yang, J., Li, S., & Yu, X. (2013). Sliding-mode control for systems with mismatched 2012, he is assistant professor with the department
uncertainties via a disturbance observer. IEEE Transactions on Industrial of Electrical Engineering at University of Isfahan. His
Electronics, 60(1), 160–169. current research interests include nonlinear observer–
Yang, J., Su, J., Li, S., & Yu, X. (2014). High-order mismatched disturbance com- controller synthesis, system identification and robust
pensation for motion control systems via a continuous dynamic sliding-mode control.
approach. IEEE Transactions on Industrial Informatics, 10(1), 604–614.