Proceedings of the IEEE

International Conference on Automation and Logistics

Qingdao, China September 2008

Virtual Compensated Sliding Mode Control of

Non-minimum Phase Nonlinear Systems with Uncertainty
CHEN Wei and REN Xia WANG Wei
College of automation and electrical engineering College of automation and electrical engineering
Qingdao University of Science and Technology Qingdao University of Science and Technology
Qingdao, 266042,China Qingdao, 266042,China
[email protected] [email protected]

Abstract -The paper studies the robust control of non- approximate algorithm, such as Yuri [2], who approximately
minimum phase nonlinear system based on the differential linearized the original system around the equilibrium point,
geometry theory. The nonlinear system could be expressed as and took the higher order item with the uncertainty as
input-output dynamics with the unstable internal modes, through perturbation of system, and introduced the virtual control to
nonlinear coordinate transformation. Virtual controller is
stabilize the internal mode dynamic and realize the output
introduced to stabilize zero dynamic, and the sliding mode
controller, which is robust to the parameter perturbations of the tracking of the original system. It is effective but can not apply
closed-loop system. Simulations show that the virtual stabilized to the system with strong nonlinearity. Tzuu presented sliding
sliding mode control scheme guarantee nice result, and is strong mode control of non-minimum phase nonlinear systems with
robust to parameter perturbations. Fuzzy theory, it was innovate idea to apply intelligent control
theory in sliding control. The robust tracking problem of
Index terms -Non-minimum phase nonlinear systems; Sliding non-minimum phase nonlinear system has also been paid more
mode control; Virtual internal mode control; Robustness attention in the application research field, both Shkolnikov [3]
I. INTRODUCTION on aerocraft and zhang [4] on shipping sailing made certain
Minimum phase nonlinear system, the zero dynamics of progress. We must improve robustness of the closed-loop
which are uniformly asymptotically stable, is investigated system to make control scheme more valuable, however,
extensively for its excellent control properties. Especially in robust control regulate and tracking problem are so difficult
recent years, it developed rapidly with the powerful nonlinear that how to make use of sliding mode control theory to realize
control techniques as feedback linearization scheme. Here, the robust control has been a new issue gradually.
zero dynamics describe the uncontrolled states where static The present study proposes an output sliding mode
state feedback is employed. It is known that a nonlinear control scheme to provide an effective control tool for
system is non-minimum phase, and it’s internal or zero non-minimum phase system with uncertainty. In present study,
dynamics are unstable, although the plant could be controlled the essential knowledge about the uncertain elements concerns
with good input-output dynamics, the closed loop control only their possible size, that is, only the sets in which the
system is hard to be stabilized. However, many physical values of uncertain quantities ranged are presumed known.
nonlinear systems, such as, the aircraft and missile control The paper is organized as follows. The problem statements
systems are typical non-minimum phase systems. The and system definition are introduced in Section 2. Switching
importance of non-minimum phase system control problem surfaces and sliding-mode control design are presented in
has been recognized recently, for which is challenging and Section 3, while Section 4 investigates the asymptotic stability
practical. of the closed-loop system and an example is illustrated in
The literatures concerning the control of non-minimum Section 5. Several conclusions are drawn at last.
phase nonlinear system mainly focus on its approximate II. PROBLEM STATEMENTS
model. These research results are investigated on the feedback
control of slightly non-minimum phase systems, which means Consider a class of affine uncertain nonlinear systems in
that the linearized system has no eigenvalue with strictly the n-dimensional smooth variable field described as
positive real parts. Hauser [1] used a minimum phase system x (t ) = f ( x,θ ) + G ( x)u
to approximate the non-minimum phase system. The (1)
decoupling matrix of this kind of system is nonsingular, but y = h( x )
near singular. Through redefining the output to increase the
Where the state variable x ∈ R n , the output and the control
relative degree, an approximate minimum phase system is
obtained, the dynamics of which is asymptotically stable variable y, u ∈ R m ; f ( x, θ ) and G ( x) are smooth variable
around the equilibrium point. However, these methods would fielding Rn ; G (0) ≠ 0 , G ( x ) = [ g1 ( x)," , g m ( x) ] and
fail if the growth rate of the zero dynamics exceeds certain
h( x) = [ h1 ( x )," , hm ( x)] ; hi (⋅) is smooth scalar function in
T
limits. In other words, these control schemes are not robust to
the uncertainty or external disturbances. R n ; θ ∈ Ω ⊂ R l denotes uncertainties of the parameters of
Most of sliding mode control research about plant. Assume that the system is independent with respect of
non-minimum phase nonlinear systems are based on the

978-1-4244-2503-7/08/$20.00 © 2008 IEEE

1643
uncertainty θ at the equilibrium point xe = 0 . (
zi = zi0 ," , ziri − 2 ) i = 1," ,m (7)
Let f 0 ( x) = f ( x, 0) and
and
Δf ( x;θ ) = f ( x;θ ) − f 0 ( x) ; ∀θ ⊂ Ω (2)
S = ( s1 ," , sm )
the nominal system of (1) is ri − 2
si ( x) = ¦ cij zij + ziri −1 i = 1," , m (8)
x = f 0 ( x) + G ( x)u (3) j =0

Assume [ r1 ," , rm ] is the relative degree vector of the ri − 2

nominal system (3) in the open subset U of R , and it’s the n

where ¦c λ
j =0
i
j
i
j
is Hurwitz polynomial. Thus, a linear

sum of relative degrees satisfy r = r1 + " rm ”n, then the coordinate transformation could be established
nonlinear system (1) could be described with its input/output "
§1 0 0 0·
and internal dynamic equations. In the coordinate of ( z ,η ) , ¨ ¸
z
§ i· ¨ 0 1 0 " 0¸
the input/output dynamic equations could be expressed as ¨ ¸ ¨ = z (9)
© si ¹ # # #¸ i
zi j = zi +j 1䯸 i = 1, 2," , rj − 1 ¨ 0 ¸
© ci ci1 " ciri −1 1¹
zrj = b ( z ,η ) + φ ( z ,η ,θ ) + a ( z ,η ) u
j
(4)
To bring the internal variables into the new coordinate system,
y j = z1 䯸 j = 1, 2," , m
j let
§η ·
ª hi ( x) º ξ =¨ ¸ (10)
« L h ( x) » ©z¹
where zi = « » , matrix a ( z ,η ) is assumed to be
f0 i

« # » In the new coordinate ( S , ξ ) , the original system (1) could be

« r1 −1 »
«¬ L f0 hi ( x) »¼ written as
nonsingular. The unstable internal dynamics are written as S = b ( S , ξ ) + φ ( S , ξ ,θ ) + a ( S , ξ ) u
(11)
η = qi ( z ,η ) (5) ξ = Q ξ + Q S +ψ ( S , ξ )
1 2

η ∈ R n − r is the state variable of the internal mode of the In which bi ( S , ξ ) denote the expression of
T
system and φ ( z ,η ,θ ) = ª¬φ1 ( z ,η , θ )," , φm ( z ,η ,θ ) º¼ describes rj − 2

bi ( z ,η ) = ¦c i
j
zij +1 +bi ( z ,η ) under the coordinate ( S , ξ ) ,
the nonlinear perturbation of the system. j =0
The internal mode of the system is linearized φ (⋅), a(⋅) and ψ (⋅) describe the new expressions of φ (⋅) ,
approximately around the equilibrium point as
a (⋅) and ψ (⋅) in the new coordinate,
η = Q1η + Q2 z + ψ ( z ,η ) (6) respectively. Q1 ∈ R ( n − m )×( n − m ) , Q2 ∈ R ( n − m )× m , since the
∂q ∂q nonsingular linear transformation does not change the property
where Q1 = ( 0, 0 )η is a known matrix, Q2 = ( 0, 0 ) z of system, the pair ( Q1 , Q2 ) keep completely controllable,
∂η ∂z
is a known matrix for which the pair (Q , Q )
1 2 is completely and the matrix Q1 still gets eigenvalues on the right half
complex plane.
controllable and ψ ( z ,η ) is the high order item of the B Design of Virtue Stabilizing Sliding Mode Control
linearized model of the system. As the non-minimum phase The virtue control would be designed to stabilize the
property of the original system, which could be explicitly seen uncontrollable internal variables, firstly, define the sliding
in this form, the matrix Q1 has eigenvalues on the right half manifold to introduce virtue control variable.
complex plane. σ = Lξ + S (12)
III. MAIN RESULTS
where S is considered as a virtual control in the sliding
A. Linear Coordinate Transformation mode, L ∈ R m× ( n − m ) is a constant matrix in which the value of
Designing a virtue control to make the uncontrollable the elements need checking. When the internal mode of the
internal modes be stabilized, consider a proper linear system reaches the designed sliding surfaces, one
coordinate transformation. Let gets S = − Lξ , then, the sliding mode equation of the system
z = ( z1 ," , zm ) (11) could be written as

1644
 ξ = (Q1 − Q2 L)ξ + ψ ( S , ξ ) (13) parameters ki are responsible for the speed with which the
sliding surfaces are reached. If ki are constants that chosen
As matrix pair ( Q1 , Q2 ) is completely controllable, one
rather large, the reaching time may be shorten, at the same
could assign the eigenvalues of (Q1 − Q2 L ) to lie sufficiently time, the high reaching speed would arise undesirable
deep in the left half plane by designing the constant matrix L . chattering. Dynamic parameters ki are chosen instead, which
Thus, the closed loop of system (13) could be asymptotically decrease the reaching speed to weaken chattering influence,
stabilized in the neighborhood of the equilibrium point. and proper large ε i are introduced to shorten the reaching time.
Sliding mode controller is designed to realize the Regulating these parameters properly could reduce the
aforementioned results, let chattering phenomenon and promote the dynamic performance
u = ueq + us (14) of the approximating mode of the system. To realize the
sliding manifold could be reached in finite time, positive real
where equivalent control u eq determinates the dynamic number ki satisfies
property of the system which reaches the sliding surfaces, ki > βi , i = 1," , m (22)
while switching control u s forces the trajectory of the
Theorem.1: consider the non-minimum phase uncertain
system to reach the sliding surfaces in finite time.
system (11), assume that ( Q1 , Q2 ) is completely controllable
Differentiating (12) with respect to time t along the
trajectory of system (11) leads to and matrix a ( z ,η ) is nonsingular, then, there exists the
following sliding mode controller, which could force the
σ = L(Q1ξ + Q2 S + ψ ( S , ξ )) + b( S , ξ ) trajectory of the system reach on the sliding surfaces and
(15)
+ φ ( S , ξ , θ ) + a( S , ξ )u remain on them, and the internal dynamics could be
asymptotically stabilized on the sliding manifolds σ = 0 .
Design equivalent controller to assure the sliding mode
asymptotical stability of the closed-loop system. u = −a(S , ξ )−1 ( LQ1ξ + LQ2 S + b(S , ξ )
(23)
ueq = −a( S , ξ )−1 ( LQ1ξ + LQ2 S + b( S , ξ )) (16) + K sgn(σ ) + Eσ )

Evidently, nonlinear perturbation has an influence on the where LQ1 ∈ R m×( n − m ) , LQ2 ∈ R m×m ; E = diag {ε i } ,
approximate mode of system. K = diag {ki } , ε i ≥ 0 , positive real number ki satisfy
σ = Lψ ( S , ξ ) + φ ( S , ξ , θ ) + a ( S , ξ )us (17) ki > βi , i = 1," , m .
Proof: As we know, the principle of the sliding mode
When θ belongs to the known gather, there must be a control is to force the trajectory of system (3) enter into the
positive variable function α i ( S , ξ ) for the uncertainty of sliding mode in finite time, i.e. satisfy the reachable condition
system, and both S and ξ satisfy the following inequation
σ iT σ i ≤ −ci σ i , ci > 0 (24)
sup { φi ( S , ξ ,θ ) } ≤ α i ( S , ξ ) i = 1," , m (18)
Substituting output sliding mode law (23) into (15), and
Around the equilibrium point of system, approximate high consider (19)
order item of system internal mode and perturbation of system σ i = −ε iσ i − ki sgn(σ i ) + γ i ( z,η ) (25)
γ ( S , ξ ) = Lψ ( S , ξ ) + φ ( S , ξ , θ ) (19) σ σ i = −ε i σ i − ki σ i + γ i ( z ,η ) σ i
T
i
2
(26)
are limitary, and their hefts satisfy Then, the following inequality is satisfied
γ i (S , ξ ) ≤ βi (20)
σ T σ ≤ −λmin ( E ) σ − ( K − γ ( z ,η ) )σ
2
where βi is a positive real number. Designing a switching (27)
controller to compensate the effect of the uncertainty and
where γ ( z,η ) = [γ 1 ( z,η )," , γ m ( z,η ) ] , ⋅
T
internal mode nonlinear high order item perturbation, the is the Euclidian
approximating technique is employed to reduce the chattering norm, λmin ( ⋅ ) denotes the smallest eigenvalue. As
phenomenon, which dues to discontinuous control and finite E = diag {ε i } and ε i ≥ 0 , the inequality λmin ( E ) ≥ 0 is satisfied.
switching times. Its principle is regulating the reaching
velocity much slower when the trajectory of the system comes Moreover, with the aid of inequality (22), one gets that the
near to the sliding mode. That is, set reachable condition (24) is satisfied. Thus, under the action of
OS control law (23), the sliding mode could be reached in
us = −a( z,η )−1 ( K sgn(σ ) + Eσ ) (21)
finite time and remain there.
{ } { }
where E = diag ε i , K = diag ki , ki > 0 and ε i ≥ 0 .The As we see from the aforementioned design of virtual
assuaged sliding mode control system, under the action of

1645
sliding control law (23), the trajectory can enter into the σ = Lξ + s (34)
sliding mode in finite time [5]. In the sliding mode manifold
of system, virtual control S disposes the sliding mode polar to where s is virtual control. Design L = ( −3 −8 ) to make the
realize the approximate stability by designing matrix K , then sliding mode has eigenvalues on the left half complex plane,
it assuages the internal mode of system (11) to control the then guarantee the system (28) to realize approximate stability
dynamic performance of out state variables. in the neighborhood of the equilibrium point. When the
IV. EXAMPLES internal mode reaches this sliding mode surface, s = ( 3 8 ) ξ ,
Consider the following uncertain nonlinear system sliding mode equation of system (28) can be rewritten as

− x3 §1 8·
ª º ª0 º ξ = ¨ ¸ξ (35)
x = « sin( x2 ) + x3 »» + ««0 »» u
« (28) © −1 −7 ¹
«¬ 2 x12 − (1 + δ ) x3 »¼ «¬1 »¼ to validate the approximate stability. According to˄23˅design
ª 0.5º sliding mode control law
x0 = ««0.2»» y = [1 0 0] x u = −2 z12 − z1 − 8η + 4 z2 + k sgn(σ ) + εσ (36)
«¬0.2»¼
where k and ε are regulating parameters, which can be chosen
Assume that the uncertainties in the system satisfy the properly in simulation to assure dynamic performance, and
inequality δ ≤ 0.2 .From calculations, one gets that the weaken the effect of buffeting. In the primary system,
relative degree of output channel is r = 2 .Choose the
equivalent control and switching control law are, respectively
nonlinear coordinate transformation as
z1 = h( x) = x1 ueq = −2 x12 − x1 − 8 x2 − 4 x3 (37)
z2 = L f h( x) = − x3 us = k sgn( x1 + 8 x2 + x3 ) + ε ( x1 + 8 x2 + x3 ) (38)
(29)
η = x2 The results of simulation are shown as follows,
where l = 0.2 and ε = 0.4 . Fig.1 and Fig.3 describe the
Under the action of nonlinear mapping, system (28) can
dynamic performance of nonlinear system internal variable,
be transformed as
where Fig .3 is responding boundary of internal variable under
z1 = z 2 the action of uncertain causes, from which we can find that
sliding mode control law realizes the internal mode assuaged
z2 = −2 z12 − (1 + δ ) z2 − u (30)
control effectively, and the dynamic performance could be got
η = sin(η ) − z2 better by regulating the parameters of sliding mode control.
Fig.2 and Fig.4 describe the responding curves of sliding
where η is the internal variable of the nonlinear system, and mode variable, where Fig .4 is responding boundary of sliding
its approximate linear model is variable under the action of uncertain causes, from which we
η = η − z2 (31) can find that sliding mode not only has strong robustness, but
also great dynamic performance.
is explicitly that the internal mode unstable, which means that
the system (28) is non-minimum phase.
To introduce virtual controller, let
s = 2 z1 + z2 (32)

In the coordinate ( s, ξ ) , system (30) and (31) could be

rewritten as

s = −2 z12 − 2 z1 + s − u
§ −2 0 · §1· (33)
ξ = ¨ ¸ξ + ¨ ¸ s
© 2 1¹ © −1¹

§z · Fig. 1 Internal variable x2 (t )

where ξ = ¨ 1 ¸ . Design sliding mode surface to make the
©η ¹
zero dynamic approximately stable under the action of virtual
control.

1646
 From the simulation, we can find that the internal mode
approximate stability under the action of sliding mode control
variable, and sliding mode variable and state variable have
strong robustness to parameter perturbation; the dynamic
performance is good according to the above chosen switch
control parameters. As Fig .5 shows, the simulated curve of
control variable has a certain buffeting, however, its effect is
so weak that the chosen parameters can satisfy system’s needs.
During the process of simulation, the design of matrix L in
sliding mode equation is quite important, which not only
satisfies the need of assuaged zero dynamic, but also
guarantees the designing requirements of dynamic property.
Fig. 2 Sliding mode variable σ (t ) V. CONCLUSIONS
The article, aimed at non-minimum phase nonlinear
systems, linearizes approximately the internal mode at the
equilibrium point. Through introducing virtual control, we can
stabilize zero dynamic. In addition, virtual control variables
are introduced into the design of sliding mode manifold to
realize that when the trajectory can enter into the sliding mode,
virtual control can assuage the internal mode and control the
dynamic performance of the closed-loop system. Designing
sliding mode controller, whose robustness can be ensured to
the parameter perturbations in the stability of the closed-loop
system. Sliding mode controller is composed of discontinuous
switch control law and continuous equivalent control law, and
the former one forces the trajectory to enter into the sliding
Fig. 3 The responding boundary of x2 (t ) under the action of uncertain causes mode in finite time, however, the later one can assuage the
system and guarantee its dynamic performance. Simulations
show that virtual stabilized sliding mode control scheme
reaches satisfied effect, and importantly has strong robustness
to parameter perturbations.
REFERENCES
[1] Hauser J. Nonlinear control via uniform system approximation[J]. 29th
control and decision conference, Honolulu, Hawaii, 1990:792-797
[2] Shkolnikov Ilya A, Shtessel Yuri B. Tracking in a class of
non-minimum-phase systems with nonlinear internal dynamics via
sliding mode control using method of system center[J]. Automatica,
2002, 38:837-842
[3] Shkonikov I.A., Aircraft non-minimum phase control in dynamics
manifolds[J].Journal of Guidance, Control, and Dynamics, 2001, Vol.
Fig .4 The responding boundary of σ (t ) under the action of uncertain causes 24(3), 566-572
[4] Zhang R.J., Chen Y.B., Sun Z.Q, Xu H.Z. Path control of a surface ship
in restricted waters using sliding mode[J]. IEEE Transactions on Control
Systems Technology. 2000, Vol.8(4), 722-732
[5] Isidor A, Krener A, Giorgi C. Nonlinear decoupling via feedback: a
differential geometric approach[J]. IEEE Trans Automatic Control, 1981,
26:331-345
[6] Shtessel Y.B. Nonlinear non-minimum phase output tracking via
dynamic sliding manifolds[J], Journal of Franklin Inst, Vol. 335B,
1998(5),841-850
[7] Lu X.Y, Spurgeon S.K. Control of nonlinear non-minimum phase
systems using dynamic sliding mode[J]. International Journal of
Systems Science, 1999, Vol. 30(2), 183-198

Fig .5 Control variable u (t )

1647