Beruflich Dokumente
Kultur Dokumente
DOI 10.1007/s11071-016-2622-7
ORIGINAL PAPER
Abstract In this paper, we are concerned with the while the tracking error are confined all times within
problem of adaptive dynamic surface error constrained the prescribed bounds. Finally, a simulation example is
control for a class of nonlinear multiple-input-multiple- provided to confirm the effectiveness of the proposed
output systems with unknown backlash-like hysteresis approach.
nonlinearities. By transforming the tracking errors into
new virtual error variables which are incorporated into Keywords Adaptive dynamic surface control ·
the proposed prescribed performance control strategy, Unknown backlash-like hysteresis · Prediction
the prescribed steady-state and transient performance error technique · Serial–parallel estimation model ·
can be ensured. Compared with the existing methods, Prescribed performance control
we introduce the prediction error which is generated
between the system state and the serial–parallel esti-
mation model to construct the adaptive laws for neural
network weights. The proposed prediction error tech- 1 Introduction
nique can be used to compensate the tracking error,
which implies that a higher accuracy of the identi- In modern society, it is already widely acknowledged
fied neural network model is achieved. It is shown that that constraints are ubiquitous in many industrial con-
the proposed control approach can guarantee that all trol systems and practical systems. If the constrained
the signals of the resulting closed-loop systems are control is not realized, it will result in unsafety speci-
bounded and the output tracks the desired trajectory, fications, physical saturation, stoppage as well as fault
and even catastrophic accidents. To handle the problem
L. Liu · Z. Wang (B) · H. Zhang
of systems without violating some constraints, many
School of Information Science and Engineering,
Northeastern University, Shenyang 110819, Liaoning, advanced control strategies were proposed, see for
China examples [1–7] and the references therein. In addition,
e-mail: zhansahn_wang@163.com several adaptive output feedback control schemes have
L. Liu also been developed such as in [8] with full-state con-
e-mail: liuleill@live.cn straints and in [9] with output constraints. This implies
H. Zhang that the tracking errors are indirectly constrained.
e-mail: hgzhang@ieee.org However, a drawback of these approaches is that the
L. Liu · Z. Wang · H. Zhang
designed controller must be redesigned to accommo-
The State Key Laboratory of Synthetical Automation for Process date the change in the BLF as mentioned in [10]. There-
Industries, Shenyang 110819, Liaoning, China fore, extra effort is required to avoid this drawback.
123
L. Liu et al.
As noticeable performance constraints, the authors [36] and so on. Although the NN-based DSC scheme
in [11] developed a performance transformation func- has been obtained much progress, few contributions
tion to ensure a prescribed performance constraint. have been achieved to the accuracy of the NN identified
Subsequently, the achievements for solving this kind models. As noted in [37], a prediction error technique
of prescribed performance constraints were published mentioned in [38,39] was incorporated into the control
[12–19]. The main idea of these results [12–18] is to framework to compensate the tracking error for SISO
introduce the so-called predefined performance bound systems. But for MIMO systems [40,41] with hystere-
(PPB) of the tracking errors to modulate the control sis, because the complete decoupling cannot be fully
performance indices (CPI). However, to date, the PPB realized, the prediction error contains information from
methodology has not been applied to the multiple- other subsystems and the non-smooth hysteresis affects
input-multiple-output (MIMO) systems with its con- the system’s normal operation, the stability analysis
trol input contains unknown backlash-like hysteresis of the closed-loop MIMO systems will become much
nonlinearities, which is more practical and important more complicated. This motivates us to address this
in industrial application. Because of the existence of paper.
coupling phenomenon among states, inputs or outputs, This paper focuses on the adaptive DSC approach
the extension of control strategies in SISO systems to with prescribed performance for a class of MIMO sys-
MIMO systems is nontrivial. tems subjected to unknown backlash-like hysteresis
In practical production and application, many actu- nonlinearities. Prediction error technique will be incor-
ators of control systems often suffer from the hys- porated into the controllers to obtain a better accu-
teresis nonlinearity [20–23]. In a mechanical trans- racy of the NN identified models. Predefined perfor-
mission system with worm gear, the worm gear is mance control (PPC) is used to make sure that the
considered as a backlash-like hysteresis. In a feed tracking error will converge to an any small resid-
drive mechanism with a lead screw drive, backlash- ual set with which a convergence rate will be no
like hysteresis is generated due to the gaps between less than a predefined value. Radial basis function
screw and nut [24]. It will lead to the problems of neural network (RBFNN) is utilized to approximate
inaccuracy, instability, and it restricts system perfor- the unknown nonlinear functions, and the DSC method
mance. A big challenge in backlash-like hysteresis is is proposed to eliminate the ’explosion of complex-
how to find out c j and b j (they are defined in Eqs. 2 ity’ problem incurred by the traditional backstepping
and 3), which are related to both the model perfor- scheme. Finally, a simulation results are provided to
mance and the hysteresis loop [25]. Recently, many confirm the effectiveness of the proposed approaches.
significant works have been achieved in the control In contrast to the large amount of existing results,
community for nonlinear systems in the presence of the main contributions of this paper are summarized as
backlash-like hysteresis, for example, see [26,27] and follows:
the reference therein. The results mentioned above (see
[21,22,26,28–30]) were all obtained based on the typ- – Different from the previous results in [27,32,34,42]
ically adaptive backstepping techniques which were where the prediction error is ignored, we develop
constructed recursively [31]. The backstepping tech- a novel adaptive backstepping DSC scheme using
nique has a main limitation as follows: the so-called prediction error technique for a class of MIMO sys-
problem of ‘explosion of complexity’ exists in the tems with unknown backlash-like hysteresis. In this
developed controllers, which is incurred by repeating case, not only the issue of ‘explosion of complex-
differentiations of nonlinear functions and virtual con- ity’ is solved, but also the accuracy of the RBFNN
trol variables. is improved.
To solve the aforementioned problem inherent in – Compared with the existing methods in [26–32,34]
the adaptive backstepping methods, several adaptive and [36], in which the PPC is not considered, the
dynamic surface control methods were proposed for control strategy presented in this paper makes sure
various systems, such as SISIO systems [19,32], that the tracking errors are constrained within the
unknown time-delay systems [27,33], stochastic non- prescribed values for all time. Besides, the require-
linear strict-feedback systems [34] and other applica- ment to construct an inverse [22] of backlash-like
tions on flight control [35] and power systems control hysteresis is avoided.
123
Adaptive dynamic surface error constrained control for MIMO systems
x̄ jn j −1 ) + d jn j −1 (x̄) D.
j (1)
⎪
⎪
⎪
⎪ + g jn −1 ( x̄ 1 , u 1 , . . . , x̄ j−1 , Remark 2 The hysteresis models can be classified into
⎪
⎪
j
⎪
⎪ u j−1 , x̄ jn j −1 )x jn j , four classes [29], which are the backlash-like hysteresis
⎪
⎪
⎪
⎪ ẋ jn j = f jn j (x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x̄ jn j )
⎪
⎪
model in [21], the conventional Prandtl–Ishlinskii (P–I)
⎪
⎪
⎪
⎪ + d jn j (x̄) + g jn j (x̄1 , u 1 , . . . , hysteresis model in [23], the generalized P–I hysteresis
⎪
⎪
⎪
⎪ model, and the modified Bouc-Wen hysteresis model
⎪
⎪ x̄ j−1 , u j−1 , x̄ jn j )u j v j ,
⎩ in [29], respectively. However, in contrast to the other
y j = x j1 , j = 1, . . . , m,
three types, the backlash-like hysteresis is more useful
where x̄ ji (k) = [x j1 (k) , . . . , x ji (k)]T ∈ R i , j = in the field of electromagnetics because the functions
1, . . . , m, i = 1, . . . , n j is the state vector of the and parameters can be fine-tuned to match experimen-
jth subsystem, y j ∈ R is the output of the jth sub- tal results [46, p. 96]. In addition, another advantage
system and x̄ = [x̄1 , x̄2 , . . . , x̄m ]T . f ji (·) , j = of backlash-like hysteresis model is that the controller
1, . . . , m, i = 1, . . . , n j are unknown nonlinear can be synthesized directly without constructing a hys-
dynamics and g ji (·) are positive control gain func- teresis inverse to mitigate effects of the hysteresis [21].
tion. d ji (x̄) is the unknown but bounded disturbance Thus, it is necessary to study the backlash-like hystere-
with d ji (X ) ≤ d̄ ji . u j (v j ) is the input vector of sis model (2).
the jth subsystem where v j represents the input of the
unknown backlash-like hysteresis in the actuator. Assumption 1 The desired trajectory y jd (t) ∈ Ω y and
its first derivatives ẏ jd (t) are smooth and known where
Remark 1 It is worth mentioning that the considered Ω y is a bounded compact set.
strict-feedback MIMO systems (1) are more general.
If the gain functions g jn j −1 (x̄1 , u 1 , . . . , x̄ j−1 , u j−1 ,
x̄ jn j −1 ) = 1 in (1), it will become the MIMO nonlin- 2.2 Prescribed performance and control objective
ear systems in [10,43]. Besides, if f jn j −1 (x̄1 , u 1 , . . . ,
x̄ j−1 , u j−1 , x̄ jn j −1 ) = f jn j −1 (x̄ jn j −1 ) and g jn j −1 According to the PPC theory [16], the designed con-
(x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x̄ jn j −1 ) = 1, the plants (1) troller should make sure the tracking error e j = y j −
will be changed into the MIMO nonlinear systems in y jd to be not beyond a specified PPB. The smooth
[42,44,45], which are studied widely in the literature. decreasing performance function ς j (t) is denoted as
[17]
According to [21], the output of the backlash-like
hysteresis in (1) is described as ς j (t) = ς0 j − ς∞ j e−l j t + ς∞ j , (4)
du j
dv j
= ξj
c j v j − u j + B j dv j , (2) where the positive constant ς∞ j stands for the maxi-
dt dt
dt mum amplitude which imposes restriction on the track-
where v j is the input of the backlash-like hysteresis. ing error to guarantee its steady state and the decreasing
ξ j and B j are unknown constants and c j > 0 is the rate e−l j t confines the convergence speed of tracking
bounded slope of the lines satisfying c j > B j . error.
123
L. Liu et al.
If the following constraint conditions hold, where the approximation error ε satisfies |ε (x)| ≤ ε̄
with the constant ε̄ > 0.
−δ j ς j (t) < e j (t) < ς j (t) if e j (0) ≥ 0, (5)
−ς j (t) < e j (t) < δ j ς j (t) if e j (0) < 0, (6) The main control objective is to develop adaptive
controllers v j , j = 1, . . . , m so that: (1) The output y j
where 0 < δ j < 1 is the prescribed scale parameter,
follows a given reference signal y jd to a small compact
then the steady-state and prescribed transient perfor-
set, (2) all the signals in the closed-loop systems are
mance of tracking error e j (t) would be guaranteed. In
bounded, and (3) the tracking error achieves prescribed
this case, the undershoot and the maximum overshoot
transient and steady-state performance.
of e j (t) are achieved by appropriately choosing the
In the following, the update laws and controllers will
value of ς j (0) and δ j ς j (0).
be developed based on Assumption 1.
Next the error transformation can be denoted as
e j (t) = ς j (t) S z j1 ,
3 RBFNN controller designs via prediction error
where S z j1 is a smooth, strictly increasing and thus
technique
invertible function and z j1 is the
transformed
error [11].
In this paper, the function S z j1 is defined as
Differentiating z j1 with regard to time, we get
M j + e−z j1 M j
S(z j1 ) = , (7) ∂ S −1 e j (t)
1 + e−z j1 ż j1 = e (t)
∂ ς jj (t) ς j (t)
where the pairs (M j , M j ) are given as
= r j f j1 x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x j1 +r j d j1 (x̄)
(−δ j , 1), if e j (0) ≥ 0, + r j g j1 x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x j1 x j2 −r j h j ,
(M j , M j ) = (8)
(−1, δ j ), if e j (0) < 0. (11)
Furthermore, we have the following inverse trans- where
formation
∂ S −1 1 ς j (M j − M j )
e j (t) rj = = ,
−1 e j (t) ς j (t) − M j e j (t) ς j (t) (e j − ς j M j )(M j ς j − e j )
z j1 = S = ln . (9) ∂ ς j (t)
ς j (t) e (t)
Mj − j ς j (t) (12)
Remark 3 Although variety of previous works and and
results are proposed to solve kinds of prescribed perfor- ς̇ j (t)e j (t)
mance constraints one after the other such as in [16–18], h j = ẏ jd + . (13)
ς j (t)
the PPB method has not been used to MIMO systems
with its control input contains unknown hysteresis non- In (11), both r j and h j contain terms relating to
linearities. The existence of backlash-like hysteresis is tracking and behavioral system properties, which are
an instability source in many systems, which means that known and they can be included in control design [18].
to develop a proper PPB scheme to compensate for or Firstly, the following transformed model can be
counteract the effect of backlash-like hysteresis is sig- obtained as
⎧
nificant. Besides, the NNs’ accuracy of approximation ⎪ ż j,1 = r j f j1 x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x j1 + r j d j1 (x̄)
⎪
⎪
in identifying and modeling is left out in [16,17]. By ⎪
⎪ + r j g j1 x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x j1 x j2 − r j h j ,
⎪
⎪
introducing prediction error technique, the accuracy of ⎪
⎪
⎪
⎪ ..
⎪
⎪ .
NNs identified models is improved in this paper. ⎪
⎪
⎪
⎨ ẋ jn j −1 = f jn j −1 x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x̄ jn j −1
Lemma 1 [16] For each continuous function f (x) ⎪ + d jn j −1 (x̄) + g jn j −1 (x̄1 , u 1 , . . . , x̄ j−1 ,
(14)
⎪
⎪
which is defined on a compact set Ω, there exists a ⎪
⎪
⎪
⎪
⎪ , x̄ jn j −1 )x jn j ,
u j−1
RBFNN (10) such that: ⎪
⎪ ẋ jn j = f jn j x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x̄ jn j + d jn j (x̄)
⎪
⎪
⎪
⎪
⎪
⎪ + g jn j x̄1 , u 1 , . . . , x̄ j−1 , u j−1 , x̄ jn j u j v j ,
sup
ωT θ (x) − f (x)
≤ ε (x) , (10) ⎩
x∈Ω y j = x j1 , j = 1, . . . , N .
123
Adaptive dynamic surface error constrained control for MIMO systems
123
L. Liu et al.
123
Adaptive dynamic surface error constrained control for MIMO systems
m m nj
1
Fig. 1 The systems diagram for the overall investigation − ρ̃ j ρ̂˙ j + τ j,i η j,i η̇ j,i . (49)
μj
j=1 j=1 i=1
ρ̂˙1 = μ1 σ1,n 1 g1,n 1 c̄1 + τ1,n 1 η1,n 1 g1,n 1 c̄1 − ρ̂1 , n j
According to (49)–(50), the terms i=1 σ̇ j,i σ j,i can
(45) be rewritten as
nj
where γ1,n 1 , τ1,n 1 , μ1 and δ1,n 1 are positive design con-
stants. σ̇ j,i σ j,i
i=1
The systems diagram for the overall investigation is
shown in Fig. 1. = σ j,1r j ω̃Tj,1 θ j,1 + r j σ j,1 ε j,1 + r j σ j,1 d j,1
+ r j σ j,1 g j,1 σ j,2 + σ j,n j ω̃Tj,n j θ j,n j + σ j,n j ε j,n j
Theorem 1 For the proposed MIMO strict-feedback
systems (1) subjected to the unknown backlash-like + σ j,n j g j,n j c j ρ̃ j − g j,n j c j σ j,n
2
j
+ σ j,n j d j,n j
hysteresis (2), under Assumptions 1, choose the vir- n j −1
tual controllers (17) and (27), parameter adaptive + σ j,i g j,i σ j,i+1 + σ j,i ε j,i − k j,i σ j,i
2
laws (25), (35) and (45)–(46), prediction errors (22), i=2
(32) and (42), and the actual controller (39), it can
+ σ j,i ω̃Tj,i θ j,i + σ j,i d j,i − 2r 2j σ j,1
2
− k j,n j σ j,n
2
.
make sure that all the signals in the MIMO nonlinear j
From the definition of the compensated tracking Based on the adaptive law (46), the following result
error (21), (31) and (41), it deduces hold
123
L. Liu et al.
1 n j −1 n j −1
ρ̃ j ρ̂˙ j = ρ̃ j σ j,n j g j,n j c̄ j − ρ̃ j ρ̂ j 1
μj σ j,i g j,i σ j,i+1 ≤ (g j,i + g j,i−1 )σ j,i
2
2
+ ρ̃ j τ j,n j η j,n j g j,n j c̄ j . (52) i=2 i=2
− g j,1 σ j,2
2
+ g j,n j −1 σ j,n
2
j
. (56)
From (24), (34) and (44), it yields
nj
Note the following facts
τ j,i η j,i η̇ j,i
i=1 ω̃Tj,i ω̂ j,i = ω̃Tj,i ω j,i − ω̃Tj,i ω̃ j,i
= r j τ j,1 η j,1 ω̃Tj,1 θ j,1 + r j τ j,1 η j,1 ε j,1 − τ j,1 1 1
= −ω̃ j,i − ω j,i 2 + ω j,i 2 . (57)
n j −1 2 4
× β j,1 η2j,1 − r j τ j,1 η j,1 h j + τ j,i η j,i ε j,i 1 1
i=2
ρ̃ j ρ̂ j = ρ̃ j ρ j − ρ̃ j ρ̃ j = −ρ̃ j − ρ j 2 + ρ j 2 .
2 4
+ τ j,i η j,i ω̃Tj,i θ j,i − τ j,i β j,i η2j,i + τ j,n j η j,n j (58)
× ε j,n j + τ j,n j η j,n j ω̃Tj,n j θ j,n j − τ j,n j β j,n j Substituting (55)–(58) into (54), and using the
× η2j,n j + τ j,n j η j,n j g j,n j c j ρ̃ j + g j,n j c j σ j,n
2
j
method of completing the square, it has
(53) ⎡
n j −1
m
V̇ ≤ ⎣−(r j σ j,1 − 1 ε j,1 )2 − A j,2 σ j,i
2
Substituting (51)–(53) into the derivative of V̇ as 2
j=1 i=3
shown in (49), it leads to
1
m − A j,1 σ j,2
2
− A j,3 σ j,n
2
j
− ρ̃ j − ρ j 2
2
V̇ = r j σ j,1 ε j,1 + r j σ j,1 d j,1 + r j σ j,1 g j,1 σ j,2
(ε j,1 − h j ) 2
j=1 − τ j,1 β j,1 η j,1 − r j
nj nj (2β j,1 )
nj
+ σ j,i × ε j,i + σ j,i d j,i
i=2 i=2
− τ j,i × β j,i (η j,i − ε j,i /(2β j,i )) + Q j
2
n j −1 n j −1 i=2
nj
+ σ j,i g j,i σ j,i+1 − k j,i 1
− δ j,i ω̃ j,i − ω j,i 2 , (59)
i=2 i=2 2
nj i=1
× σ j,i
2
+ δ j,i ω̃Tj,i ω̂ j,i + ρ̃ j ρ̂ j where
i=1
+ r j τ j,1 η j.1 ε j,1 − r j × τ j,1 η j.1 h j 1 1
A j,1 = k j,2 − g 2j,1 − g j,2 − 1,
nj nj 2 2
1 1
− τ j,i β j,i η2j,i + τ j,i η j,i ε j,i A j,2 = k j,i − g j,i − g j,i−1 − 1,
i=1 i=2
2 2
1
− 2r 2j σ j,1
2
− k j,n j σ j,n
2
. (54) A j,3 = k j,n j − g j,n j −1 − 1,
j 2
nj nj nj
By using the Young’s inequality, one obtains 1 1 2 1
Qj = d 2j,i M + ε j,i M + δ j,i ω j,i 2
1 2 1 2 2 4
i=1 i=1 i=1
σ j,i ε j,i ≤ σ + ε2 ,
2 j,i 2 j,i M 1
1 1 2 + ρ j 2 + τ j,1 r 2j (ε j,1M − h j )2 /(4β j,1 )
r j σ j,1 d j,1 ≤ r 2j σ j,1
2
+ dM , 4
2 2 nj
1
1 1 + τ j,i ε2j,i M /β j,i .
r j σ j,1 g j,1 σ j,2 ≤ r 2j σ j,1
2
+ g 2j,1 σ j,2
2
, 4
2 2 i=2
nj nj nj
1 2 1 2
σ j,i d j,i ≤ σ j,i + dM , (55) Choose the design parameters k j,2 , k j,i and k j,n j
2 2 such that
i=2 i=2 i=2
123
Adaptive dynamic surface error constrained control for MIMO systems
k j,i ≥ g j,i + g j,i−1 + 1,
2 2
1 Remark 5 A mass of existing backstepping-based DSC
k j,n j −1 ≥ g j,n j −1 + 1, results have been also proposed for SISO systems [32],
2
Then, it implies that A j,1 ≥ 0, A j,2 ≥ 0 and A j,3 ≥ MIMO systems [42], stochastic systems [34] and time-
0. Therefore, if at least one of the following inequalities delay systems [27] in strict-feedback forms, but these
hold contributions have been achieved without emphasizing
the accuracy of the NN or fuzzy logic system (FLS)
σ j,2 ≥ Q j /A j,1 ,
identified models. In this paper, by introducing the
n j −1
serial–parallel estimation models as shown in (22), (32)
A j,2 σ j,i
2
≥ Q j, and (42), the prediction errors generated between the
i=3
system states and those models are incorporated into
or the design procedure to achieve faster adaption ability.
σ j,n j ≥ Q j /A j,3 ,
Remark 6 In contrast to the existing results in related
it can be concluded that V̇ ≤ 0. filed, the proposed control method has the following
Thus, based on the Lyapunov stability analysis the- advantages:
orem, compensated tracking errors σ ji , the prediction
(a) A restriction in [27,32,34,42,43] is that the
errors η ji and the adaptive laws ω̂ ji and ρ̂ j are bounded.
backlash-like hysteresis (a non-smooth input) is
As stated in [17], owing to the properties of the transfor-
neglected since the backlash-like hysteresis is a
mation, r j > 0 and r j , h j remain bounded. From the
common phenomenon in engineering. Here, a
serial–parallel estimation model, it can be concluded
compensatory adaptive law (38) is introduced to
that x̂˙ ji are bounded. Besides, consider (24), (34) and
offset this non-smooth input, which is different
(44), η̇ ji are bounded, which further leads to the bound-
from the previous results in [21–23] that an inverse
edness of ż ji . This implies that the systems states are
of hysteresis is needed.
bounded, which can result in the boundedness of the
(b) In the work of [21–23,26–32,34–36,49], the pre-
virtual control. Based on (17), (27) and (37), it can
scribed bound of tracking error is ignored. On
be seen that s ji are bounded. Hence, z 11 and x cji are
this occasion, even though the steady-state per-
bounded. According to the prediction errors in (22),
formance is realized, the transient performance
(32) and (42), the boundedness of x̂ ji can be guaran-
cannot be guaranteed. In this paper, the proposed
teed. Some manipulations can ensure that the signals
approach makes sure that the tracking error is con-
a ji , ȧ ji and ṡ ji are also bounded. Therefore, all the sig-
strained within the prescribed values for all time,
nals in the MIMO systems are bounded. This completes
which means that both steady-state performance
the proof.
and transient performance are obtained.
To summarize the overall design procedure, the fol-
lowing complete algorithm is provided: Remark 7 In a practical application, the tracking error
(1) Select appropriately basis function vectors of may exceed a certain prescribed threshold. The practi-
neural networks, and construct the NN approxi- cal implication of this paper is that the presented predic-
mator. tion error technique-based PPC method could avoid the
(2) Set the initial values for x ji (0), ω ji (0) and ρ j (0), aforementioned problem. In addition, only the informa-
and choose proper design parameters (positive) tion of desired trajectory y jd and its first-order deriv-
k ji , α ji , β ji , γ ji , τ ji , μ j , and δ ji . ative ẏ jd is required in this study while y jd , ẏ jd up to
(n )
(3) Specify a positive parameter ζ j0 , select the min- n j th order derivatives ẏ jd j is needed in the traditional
imum allowable convergence rate l j and set the DSC methods such as in [27,34,42,43]. This relax-
maximum admissible amplitude ζ j∞ . ation could reduce the computational burden when the
(4) Calculate the adaptive parameter estimate values developed approach is applied in a practical system.
in (24), (34), (44) and (45) by utilizing (21), (31) Therefore, this paper provides a theoretical framework
and (41). for the practical application.
123
L. Liu et al.
1
Remark 8 From the definition of (5) and (6), for all x
t ≥ 0, the tracking error e j (t) stays in the constraint
11
0.8 y
1d
condition. This implies that the initial value of tracking 0.6
error cannot exceed the PPB. In addition, at the initial 0.4
instant t = 0, the maximum undershoot is less than
0.2
ζ j (0) and can be controlled by selecting ζ j (0) close
enough to |e j (0)|. Therefore, proper choice of the per- 0
−0.6
−0.8
4 Simulation studies
−1
0 5 10 15
To validate the effectiveness of the proposed control seconds
method, we consider the following MIMO systems
with 3 subsystems subjected to backlash-like hysteresis Fig. 2 The curves of x11 (t) and yd1 (t). (Color figure online)
characteristic
⎧
⎨ ẋ11 = f 11 (x11 ) + g11 (x11 ) x12 ,
⎪ 2
x
21
ẋ12 = f 12 (x11 , x12 ) + g12 (x11 , x12 ) u 1 (v1 ) + d1 , y2d
⎪
⎩ 1.5
y1 = x11 ,
⎧ 1
⎨ ẋ21 = f 21 (·) + g21 (·) x22 ,
⎪
ẋ22 = f 22 (·) + g22 (·) u 2 (v2 ) + d2 , 0.5
⎪
⎩
y2 = x21 ,
⎧ 0
c j v j − u j + B j dv j , 0.5, x2,2 (0) = −0.5, x3,1 (0) = 0.5, x3,2 (0) = 1,
dt dt
dt θ̂1,2 (0) = 0.2I , θ̂2,2 (0) = 0.1I , θ̂3,2 (0) = 0.5I ,
where v j are the input of the hysteresis actuators, c j = ρ̂1 (0) = 0.1, ρ̂2 (0) = 0, ρ̂3 (0) = 0.4. All the
3.1635, ξ j = 5, B j = 0.345. parameters are chosen as ς0,1 = 9.3, ς∞,1 = 0.35,
The reference output are assumed to be: y1d = 0.5× l1 = 2, ς0,2 = 9.5, ς∞,2 = 0.55, l2 = 2, ς0,3 = 1.6,
sin 2π t, y2d = 0.8 sin π t and y3d = sin π t + cos π t. ς∞,3 = 0.1, l1 = 1, δ1 = 0.8, δ2 = 0.85, δ3 = 0.8,
The main objective of this paper is to construct con- k11 = 15, k12 = 15, k21 = 12, k22 = 12, k31 = 20,
trol laws v j , j = 1, 2, 3 such that the system outputs y j k32 = 20, α1 = 0.005, α2 = 0.005, α3 = 0.01, γ1 = 3,
can track a reference input y jd as well as the following γ2 = 3, γ3 = 1, τ1 = 1, τ2 = 1, τ3 = 1, μ1 = 1,
123
Adaptive dynamic surface error constrained control for MIMO systems
2.5 10
y3d e2
2 x31 8 -- δ2 ς2
ς2
1.5 6
1 4
0.5 2
0 0
−0.5
−2
−1
−4
−1.5
−6
−2
0 5 10 15 −8
0 5 10 15
seconds
seconds
Fig. 4 The curves of x31 (t) and yd3 (t). (Color figure online)
Fig. 6 The trajectories of tracking error e2 and the prescribed
error bounds. (Color figure online)
10
e1
2
8 -- δ1ς1
ς1
6 1.5
4 1
ς
3
2
0.5
0
0
−2
−4 −0.5
e3 in this paper −ς δ
3 3
−6 −1
−8
0 5 10 15 −1.5
0 5 10 15
seconds
Fig. 7 The trajectories of tracking error e3 and the prescribed
Fig. 5 The trajectories of tracking error e1 and the prescribed error bounds. (Color figure online)
error bounds. (Color figure online)
123
L. Liu et al.
5 control input v1 10
e without PPC in [39]
2
0 8 performance bound (lower)
−5 performance bound (upper)
0 5 10 15 6
seconds
4
5
control input v
2
0 2
−5
0
0 5 10 15
−2
seconds
5 −4
control input v3
0
−6
−5
0 5 10 15 −8
0 5 10 15
seconds seconds
Fig. 8 The trajectories of actual controllers v1 v2 and v3 . (Color Fig. 10 The curves of tracking error e2 (green lines) without
figure online) prescribed performance in [43]. (Color figure online)
10 2
e without PPC in [39]
1 performance bound (lower)
8 performance bound (lower) performance bound (upper)
performance bound (upper) 1.5 e without PPC in [39]
3
6
1
4
2 0.5
0 0
−2
−0.5
−4
−1
−6
−8 −1.5
0 5 10 15 0 5 10 15
seconds
seconds
Fig. 11 The curves of tracking error e3 (green lines) without
Fig. 9 The curves of tracking error e1 (green lines) without
prescribed performance in [43]. (Color figure online)
prescribed performance in [43]. (Color figure online)
123
Adaptive dynamic surface error constrained control for MIMO systems
tracking errors. The unknown system dynamics have 10. Tong, S.C., Sui, S., Li, Y.M.: Fuzzy adaptive output feedback
been estimated by RBFNNs. In contrast with the pre- control of MIMO nonlinear systems with partial tracking
errors constrained. IEEE Trans. Fuzzy Syst. doi:10.1109/
vious results, the prediction errors which are generated TFUZZ.2014.2327987
between the system states and the serial–parallel esti- 11. Benchlioulis, C.P., Rovithakis, G.A.: Robust partial-state
mation models have been incorporated into the devel- feedback prescribed performance control of cascade systems
oped adaptive laws for RBFNNs which can improve the with unknown nonlinearities. IEEE Trans. Autom. Control
56, 2224–2239 (2011)
accuracy of the identified RBFNNs model. An impor- 12. Marantos, P., Eqtami, A., Bechlioulis, C.P., Kyriakopoulos,
tant contribution is to introduce the PPC methods for K.J.: A prescribed performance robust nonlinear model pre-
the more general MIMO systems with hysteresis to dictive control framework. In: European IEEE Control Con-
make sure that the tracking errors do not exceed the ference (ECC), pp. 2182–2187 (2014)
13. Su, X., Shi, P., Wu, L., Karimi, H.R.: Design on fuzzy con-
prescribed bounds. It is shown that the proposed con- trol for a class of stochastic nonlinear systems. In: IEEE
trol strategies can guarantee that all the signals of the American Control Conference (ACC), pp. 544–547 (2014)
resulting closed-loop systems are bounded and the out- 14. Song, H.T., Zhang, T., Zhang, G.L., Lu, C.J.: Robust
puts can track the desired trajectories, while the track- dynamic surface control of nonlinear systems with pre-
scribed performance. Nonlinear Dyn. 76, 599–608 (2014)
ing errors are confined all times within the prescribed 15. Wang, W., Wen, C.: Adaptive actuator failure compensa-
bounds. tion control of uncertain nonlinear systems with guaranteed
transient performance. Automatica 46, 2082–2091 (2009)
Acknowledgments This work was supported by the National 16. Benchlioulis, C.P., Rovithakis, G.A.: Neuro-adaptive
Natural Science Foundation of China of Grants 61473070 force/position control with prescribed performance and
and 61433004, the Fundamental Research Funds for the Cen- guaranteed contact maintenance. IEEE Trans. Neural Netw.
tral Universities (Grant Nos. N130504002, N140406001 and 21, 1857–1868 (2010)
N130104001), and SAPI Fundamental Research Funds (Grant 17. Benchlioulis, C.P., Rovithakis, G.A.: Adaptive control with
No. 2013ZCX01). guaranteed transient and steady state tracking error bounds
for strict feedback systems. Automatica 45, 532–538 (2009)
18. Bechlioulis, C.P., Doulgeri, Z., Rovithakis, G.A.: Guaran-
teeing prescribed performance and contact maintenance via
an approximation free robot force/position controller. Auto-
References matica 48, 360–365 (2012)
19. Liu, Y.J., Tong, S.C.: Barrier Lyapunov functions-based
1. How, B.V.E., Ge, S.S., Choo, Y.S.: Control of coupled vessel, adaptive control for a class of nonlinear pure-feedback
crane, cable, and payload dynamics for subsea installation systems with full state constraints. Automatica (in press).
operation. IEEE Trans. Control Syst. Technol. 19, 208–220 doi:10.1016/j.automatica.2015.10.034
(2011) 20. Zhang, X.Y., Lin, Y.: Adaptive tracking control for a class
2. Tee, K.P., Ge, S.S., Tay, E.H.: Adaptive control of electro- of pure-feedback nonlinear systems including actuator hys-
static microactuator with bidirectional drive. IEEE Trans. teresis and dynamic uncertainties. IET Control Theory Appl.
Control Syst. Technol. 17, 340–352 (2009) 5, 1868–1880 (2011)
3. Kothare, M.V., Balakrishnan, V., Morari, M.: Robust 21. Su, C.Y., Stepanenko, Y., Svoboda, J., Leung, T.P.: Robust
constrained model predictive control using linear matrix adaptive control of a class of nonlinear systems with
inequalities. Automatica 32, 1361–1379 (1996) unknown backlash-like hysteresi. IEEE Trans. Autom. Con-
4. Hu, T., Lin, Z.: Control Systems with Actuator Saturation: trol 45, 2427–2432 (2000)
Analysis and Design. Birkhuser, Boston (2001) 22. Tao, G., Petar, V.K.: Adaptive control of plants with
5. Tee, K.P., Ge, S.S., Tay, E.H.: Barrier Lyapunov functions for unknown hystereses. IEEE Trans. Autom. Control 40, 200–
the control of output-constrained nonlinear systems. Auto- 212 (1995)
matica 45, 918–927 (2009) 23. Su, C.Y., Wang, Q.Q., Chen, X.K., Rakheja, S.: Adaptive
6. Niu, B., Zhao, J.: Barrier Lyapunov functions for the output variable structure control of a class of nonlinear systems with
tracking control of constrained nonlinear switched systems. unknown Prandtl–Ishlinskii hysteresis. IEEE Trans. Autom.
Syst. Control Lett. 62, 963–971 (2013) Control 50, 2069–2073 (2005)
7. Liu, L., Liu, Y.J., Chen, C.L.P.: Adaptive neural network 24. Dong, R.L, Tan, Y.H., Janschek, K.: Non-smooth predic-
control for a DC motor system with dead-zone. Nonlinear tive control for wiener systems with backlash-like hystere-
Dyn. 72(1–2), 141–147 (2013) sis. IEEE/ASME Trans. Mechatron. doi:10.1109/TMECH.
8. Liu, Y.J., Li, D.J., Tong, S.C.: Adaptive output feedback 2015.2453794
control for a class of nonlinear systems with full-state con- 25. Hassani, V., Tjahjowidodo, T., Do, T.N.: A survey on hys-
straints. Int. J. Control 87, 281–290 (2014) teresis modeling, identification and control. Mech. Syst. Sig-
9. Ren, B.B., Ge, S.S., Tee, K.P., Lee, T.H.: Adaptive neural nal Process. 49, 209–233 (2014)
control for output feedback nonlinear systems using a barrier 26. Wang, H., Chen, B., Liu, K., Liu, X., Lin, C.: Adaptive neural
Lyapunov function. IEEE Trans. Neural Netw. 21, 1339– tracking control for a class of nonstrict-feedback stochastic
1344 (2010)
123
L. Liu et al.
nonlinear systems with unknown backlash-like hysteresis. 39. Pan, Y., Er, M.J., Sun, T.: Composite adaptive fuzzy control
IEEE Trans. Neural Netw. Learn. Syst. 25, 947–958 (2014) for synchronizing generalized Lorenz systems. Chaos: an
27. Wang, M., Liu, X., Shi, P.: Adaptive neural control of pure- interdisciplinary. J. Nonlinear Sci. 22, 023144 (2012)
feedback nonlinear time-delay systems via dynamic surface 40. Chen, C., Liu, Z., Zhang, Y., Chen, C.L.P., Xie, S.: Adaptive
technique. IEEE Trans. Syst. Man Cybern. Part B: Cybern. control of MIMO mechanical systems with unknown actu-
41, 1681–1692 (2011) ator nonlinearities based on the Nussbaum gain approach.
28. Liu, L., Wang, Z.S., Zhang, H.G.: Adaptive NN fault- IEEE/CAA J. Autom. Sin. 3(1), 26–34 (2016)
tolerant control for discrete-time systems in triangular forms 41. Chen, L., Wang, Q.: Adaptive robust control for a class of
with actuator fault. Neurocomputing 152, 209–221 (2015) uncertain MIMO non-affine nonlinear systems. IEEE/CAA
29. Liu, Z., Lai, G.Y., Zhang, Y., Chen, X., Chen, C.L.P.: Adap- J. Autom. Sin. 3(1), 105–112 (2016)
tive neural control for a class of nonlinear time-varying delay 42. Han, S.I., Lee, J.: Partial tracking error constrained fuzzy
systems with unknown hysteresis. IEEE Trans. Neural Netw. dynamic surface control for a strict feedback nonlinear
Learn. Syst. 25, 2129–2140 (2014) dynamic system. IEEE Trans. Fuzzy Syst. 22, 1049–1061
30. Huang, J., Dou, L.H., Fang, H., Chen, J., Yang, Q.K.: Distrib- (2014)
uted backstepping-based adaptive fuzzy control of multiple 43. Tong, S.C., Li, Y.M., Feng, G., Li, T.S.: Observer-based
high-order nonlinear dynamics. Nonlinear Dyn. 81, 63–75 adaptive fuzzy backstepping dynamic surface control for a
(2015) class of MIMO nonlinear systems. IEEE Trans. Syst. Man
31. Seto, D., Annaswamy, A., Baillieul, J.: Adaptive control of Cybern. Part B: Cybern. 41, 1124–1135 (2011)
nonlinear systems with a triangular structure. IEEE Trans. 44. Wang, Z.S., Liu, L., Zhang, H.G., Xiao, G.Y.: Fault-tolerant
Autom. Control 39, 1411–1428 (1994) controller design for a class of nonlinear MIMO discrete-
32. Wang, D., Huang, J.: Neural network-based adaptive time systems via online reinforcement learning algorithm.
dynamic surface control for a class of uncertain nonlinear IEEE Trans. Syst. Man Cybern.: Syst. (in press). doi:10.
systems in strict-feedback form. IEEE Trans. Neural Netw. 1109/TSMC.2015.2478885
16, 195–202 (2005) 45. Liu, L., Wang, Z.S., Zhang, H.G.: Adaptive fault-tolerant
33. Xie, X.J., Zhao, C.R.: State feedback stabilization of sto- tracking control for MIMO discrete-time systems via rein-
chastic feedforward nonlinear systems with input time- forcement learning algorithm with less learning parameters.
delay. Acta Autom. Sin. 40(12), 2972–2976 (2014) IEEE Trans. Autom. Sci. Eng. (in press). doi:10.1109/TASE.
34. Chen, W., Jiao, L., Li, J., Li, R.: Adaptive NN backstep- 2016.2517155
ping output-feedback control for stochastic nonlinear strict- 46. Macki, J.W., Nistri, P., Zecca, P.: Mathematical models for
feedback systems with time-varying delays. IEEE Trans. hysteresis. SIAM Rev. 35, 94–123 (1993)
Syst. Man Cybern. Part B: Cybern. 40, 939–950 (2010) 47. Antonelli, R., Astolfi, A.: Continuous stirred tank reactors:
35. Xu, B., Huang, X., Wang, D., Sun, F.: Dynamic surface con- easy to stabilise. Automatica 39, 1817–1827 (2003)
trol of constrained hypersonic flight models with parameter 48. Kaddissi, C., Kenne, J.P.: Maarouf Saad: indirect adaptive
estimation and actuator compensation. Asian J. Control 16, control of an electrohydraulic servo system based on non-
162–174 (2014) linear backstepping. IEEE/ASME Trans. Mechatron. 16,
36. Mehraeen, S., Jagannathan, S., Crow, M.: Power system sta- 1171–1177 (2011)
bilization using adaptive neural network-based dynamic sur- 49. Wang, H.Q., Liu, X.P., Chen, B., Zhou, Q.: Adaptive fuzzy
face control. IEEE Trans. Power Syst. 26, 669–680 (2011) decentralized control for a class of pure-feedback large-scale
37. Xu, B., Shi, Z.K., Yang, C.G., Sun, F.C.: Composite neural nonlinear systems. Nonlinear Dyn. 75(3), 449–460 (2014)
dynamic surface control of a class of uncertain nonlinear
systems in strict-feedback form. IEEE Trans. Neural Netw.
44, 2626–2634 (2014)
38. Bellomo, D., Naso, D., Turchiano, B., Babu ska, R.: Com-
posite adaptive fuzzy control. In: Proceedings of the 16th
IFAC World Congress Prague, Czech Republic, pp. 97–102
(2005)
123