2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control

Mexico City, Mexico. September 26-28, 2012

Practical Stability of Control Processes Governed

by Semi-explicit DAEs
R. Juárez1,∗ , V. Azhmyakov1 and A. Poznyak1
1
Department of Automatic Control. CINVESTAV
Av. Instituto Politécnico Nacional 2508. 07300 México D.F., México
Phone (5255) 57473795 Fax (5255) 57473982 E-mail:rjuarez@ctrl.cinvestav.mx
∗
Corresponding author.

Abstract— This paper deals with a new approach of the method we propose is illustrated by means of an
to robust control design for a class of nonlinearly example of a strictly nonlinear implicit control system.
affine control systems. The dynamical models under The remainder of our paper is organized as follows.
consideration are described by a special class of struc-
tured implicit differential equations called semi-explicit Section 2 contains the problem formulation and related
differential-algebraic equations (of index one), in the concepts. The basic analytic results related to the AE
presence of additive bounded uncertainties. The pro- method are collected in Section 3. In this part of the
posed robust feedback design procedure is based on manuscript we also discusses a robust feedback control
an extended version of the classical invariant ellipsoid design procedure based on a ”minimal-size” attractive el-
technique that we call the Attractive Ellipsoid (AE)
method. The theoretic schemes elaborated in our con- lipsoid. Section 4 is devoted to the corresponding computa-
tribution are illustrated by a simple computational tional technique that extends the elaborated methodology.
example. In fact, we reduce the robust controller design problems
Keywords: semi-explicit DAE, implicit systems, robust to an auxiliary LMI-constrained optimization problem.
control, attractive ellipsoid, descriptor method. Section 5 summarizes the paper.
I. Introduction II. Problem Formulation and Some
Recently the interest to the new powerful robust control Preliminaries
design approaches to the non-conventional control systems Let us consider the following initial value problem for
has been significantly increased (see e.g. [2], [4], [8], [9], the nonlinear implicit system with an affine structure
[11], [12], [13], [15], [17], [18]). For example, the robust
E ẋ(t) = f (x(t)) + Bu(x(t)) + η(t), t ∈ R+
stabilizing of dynamic models described by implicit differ- (1)
ential equations has emerged as a challenging problem of x(0) = x0 ,
the modern systems theory [5], [10], [20]. In our paper, we where E ∈ Rn×n is a given matrix with det{E} = 0 and
consider a particular family of control processes governed B ∈ Rn×m . By x(t) ∈ Rn , t ∈ R+ and u(x) ∈ Rm , x ∈ Rn
by so-called semi-explicit differential algebraic equations we denote here the n-dimensional state vector and the
(DAEs) of index one in the presence of bounded uncer- m-dimensional feedback-type control vector, respectively.
tainties. We restrict our consideration to the mathematical Problem (1) constitutes an adequate modeling framework
models of nonlinear systems with affine structures. Our for a wide class of control systems characterized by an
aim is to elaborate a constructive control design scheme implicit state equation in presence of some bounded un-
that guarantees the practical stability (stabilization with certainties η(t) ∈ Rn , t ∈ R+ , where ||η(t)||Qη ≤ 1
respect to a bounded ”small” region) of the closed-loop (see [6]). By || · ||H , where H is a n × n-dimensional
system realizations. The concrete robust control synthesis symmetric positive defined√matrix, we denote a weighted
approach proposed in our contribution is a consequence of Euclidean norm ||x||H := xT Hx. Let us now make the
the newly established ”Attractive Ellipsoid” (AE) method basic technical assumptions associated with (1):
established in [1], [11], [12], [14], [16], [15], [18]. n n
• f : R → R is a continuously differentiable function
In our paper, we investigate a particular family of dy-
with a bounded derivative and moreover,
namic processes characterized by the implicit differential
equations with the so called quasi-Lipschitz right-hand ||f (x) − Ax||2Qf ≤ δ + h||x||2Qx ∀ x ∈ Rn , h ∈ R+ ;
sides that can be reduced to the semi-explicit DAEs of
index one (see e.g., [10] for details). The main mathemat- • η(·) is assumed to be a bounded measurable function;
ical tool used in our paper is an appropriate extension of • u(x) is chosen in the form of a linear feedback, namely,
the classic Lyapunov-based techniques (consult [2], [8], [9], u(x) = Kx, where K ∈ Rm×n .
[17], [21]). However, we apply these techniques in combi- Here, Qf and Qx are symmetric positive defined matrices.
nation with the descriptor method [7]. The effectiveness Through this paper, functions f (·) that satisfy the first

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 31
978-1-4673-2169-3/12/$31.00 ©2012 IEEE
 2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control
Mexico City, Mexico. September 26-28, 2012

basic hypothesis are called quasi-Lipschitz functions (the Theorem 1 gives a rise to a further re-transformation of
first hypothesis from the above main conditions). The the original closed-loop system (2). Let us additionally as-
simple one-dimensional illustration of the quasi-Lipschitz sume that the given matrix pair (E, A) is regular, namely,
restrictions is given on Fig. 1. 
I 0
 
J 0

(E, A) ∼ ,
0 N 0 I
for some nonsingular matrices Π and Ψ. Introduce a
coordinate transformation given by x = Ψz and multiply
the both sides of equation in (2) by Π (from the left), we
now rewrite (2) as a system

ż1 (t) = Jz1 (t) + Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t)),

N ż2 (t) = z2 (t) + Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t)),
z(0) = Ψ−1 x0 .
(3)
T
where z := (z1 , z2 ) such that dim{z1 } = dim{J} and
dim{z2 } = n− dim{z1 }. Assume that the nilpotent matrix
N in (3) has the nilpotency index ν. In that case the second
Fig. 1: The one-dimensional domain of quasi-Lipschitz func- equation in (3) implies the following algebraic equation
tions.
N ν−1 z2 (t)+
We now use the basic assumption and rewrite (1) in the N ν−1 Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t)) = 0.
following equivalent form Finally, we obtain a system of so-called semi-explicit dif-
E ẋ(t) = Ax(t) + (f (x(t)) − Ax(t) + BKx(t) + η(t)), ferential algebraic equations (DAEs)
(2)
x(0) = x0 , t ∈ R+ , ż1 (t) =Jz1 (t) + Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t)),
0 =G(z1 , z2 , t), z(0) = Ψ−1 x0 .
Let us introduce some useful concepts and facts associated
(4)
with a matrix pair (Γ1 , Γ2 ), where Γ1 , Γ2 ∈ Rn×n . We refer
ν−1 ν−1
to [10] for the additional technical details. where G(z1 , z2 , t) := N z2 (t) + N Π(f (Ψz(t)) −
AΨz(t) + BKΨz(t) + η(t)), (4) is equivalent to the origi-
Definition 1: A matrix pair (Γ1 , Γ2 ) is called regular if nally given initial value problem (2).
the characteristic polynomial p(λ) := det(λΓ1 −Γ2 ), λ ∈ R
is not a zero polynomial. A matrix pair which is not regular Definition 3: [10] A DAE of the form
is called singular. ẋ = g1 (x, y, t) 0 = g2 (x, y, t)
Definition 2: Two pairs of matrices (Γ1 , Γ2 ) and is called semi-explicit DAE. Where x ∈ Rn , y ∈ Rm , are
(Σ1 , Σ2 ), Σ1 , Σ2 ∈ Rn×n , are called strongly equivalent dependent variables, g1 (·, ·, ·), g2 (·, ·, ·) are continuously
if there exist nonsingular matrices Θ, ∆ ∈ Rn×n such that differentiable functions.
In our paper we also restrict our consideration to im-
Σ1 = ΘΓ1 ∆, Σ2 = ΘΓ2 ∆
plicit systems (2) which can be transformed, via matrix
if this is the case, we write (Γ1 , Γ2 ) ∼ (Σ1 , Σ2 ) transformation, into a semi-explicit DAE of the form (4)
with differential index one.
Recall that a matrix N is called a nilpotent matrix of Recall that a differentiation index of a DAE (4) is a
the index ν ∈ N if N ν = 0 and N ν−k 6= 0 for all 1 ≤ k ≤ most minimal number L ∈ N such that the equation
ν − 1, k ∈ N. Using the above concepts, we are now ready
to discuss so called Weierstrass canonical form (see e.g., d(l)
(N ν−1 z2 (t)+
[10]). dt(l)
N ν−1 Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t))) = 0,
Theorem 1: Let (Γ1 , Γ2 ) be regular. Then, we have
where l = 1, ..., L, implies an semi-explicit DAE with
respect to z2 (see [10] for details). Equation above has
   
I 0 J 0
(Γ1 , Γ2 ) ∼ , , the next form for index one:
0 N 0 I
∂ ∂ ∂
where J is a matrix in Jordan canonical form and N is a (G(z1 , z2 , t)) = (G(z1 , z2 , t))ż1 + (G(z1 , z2 , t))ż2
∂t ∂z1 ∂z2
nilpotent matrix also in Jordan canonical form. Moreover, ∂
it is allowed that the one or the other block is not present. + (G(z1 , z2 , t)),
∂t

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 32
978-1-4673-2169-3/12/$31.00 ©2012 IEEE
 2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control
Mexico City, Mexico. September 26-28, 2012

The index one property requires that G(z1 , z2 , t) be solv- From the point of view of an implementable control
able with respect to z2 . In other words, the differentiation applications we are usually interested to construct an
index is one if by differentiation on time of the algebraic attractive ellipsoid EP that possesses a minimal size (in
part of an implicit differential system, results the system: an adequate sense). This requirement can be formalized in
the form of a specific minimization problem subject to the
ż1 (t) =Jz1 (t) + Π(f (Ψz(t)) − AΨz(t) + BKΨz(t) + η(t)),
characteristic parameters of EP . This minimality property
∂ ∂
0= (G(z1 , z2 , t))ż1 + (G(z1 , z2 , t))ż2 can be formalized by the following minimization problem
∂z1 ∂z2
∂ minimize tr{P −1 }
+ (G(z1 , z2 , t)), (6)
∂t subject to P > 0, P = P T , K ∈ Υ,
(5)
where Υ ⊂ Rm×n is the set of matrices in which ensure in-
det( ∂z∂ 2 (G(z1 , z2 , t)))
which is solvable for (ż1 , ż2 ) if 6= 0. variance of attractive ellipsoid EP . The basic optimization
The above assumptions on the semi-explicit DAE (4) of problem (6) determines a suitable control function of the
differential index one or on the initial implicit system (2) linear feedback type that guarantees asymptotical stability
makes it possible to use all the conventional systematic and the invariance of the designed ellipsoidal region EP
concepts associated with the dynamic behavior of the (called attractive ellipsoid). Evidently, next we need to
implicit systems of the type (2) including all the classic give a constructive characterization of the set Υ and
Lyapunov stability and the set stability concepts. propose an implementable method for a computational
Let us now recall these standard definitions. A set D in treatment of the optimization problem (6). The above-
the state space of system of a dynamic system is said to be mentioned concrete specification of the set Υ will be
(positively) invariant if an admissible trajectory initiated proceeded (in the next sections) by means of a Lyapunov-
in this set remains inside the set at all future time instants. like analysis and using the technique of the linear matrix
Let us denote by Ω(z(0)) the (positive) limit set of (4) (the inequalities (see [3] for LMIs).
set of all positive limit points. We refer to [2], [8], [9] for III. The Attractive Ellipsoid Method
further theoretical background and useful results. As mentioned in Introduction, the basic results from
Definition 4: A compact invariant set D ⊂ Rn of the the Lyapunov set stability theory can not be directly
closed-loop semi-explicit system (4) of index one is called applied to the implicit nonlinear system (2). Moreover,
asymptotically Lyapunov stable if, Ω(z(0)) ∈ D and the analogous theory for the general implicit systems (see
• for all ǫ > 0 there exists δ1 > 0 such that
e.g., [5], [10], [19]) is quite sophisticated in the context of
dist{z(0), D} < δ1 implies that dist{z(t), D} < ǫ, t ≥ practical implementations. Motivated from that fact, our
0 (the Lyapunov stability of the set); study of the asymptotical stability properties of E will be
• there exists δ2 > 0 such that dist{z(0), D} < δ2
based on the following auxiliary result.
implies limt→∞ dist{z(t), D} = 0 Lemma 1: Let a function V : R+ → R+ satisfies the
(the attraction of the set). following differential inequality: V̇(t) ≤ −αV(t) + β, where
Here dist{z, D} := minz̃∈D ||z−z̃||Rn is the usual Euclidean α > 0 and β > 0 are some constants. Then limt↑∞ V(t) ≤
distance between a point z ∈ Rn and D. β/α and moreover,
Our aim is to specify an invariant set for the closed- β β
V(t) ≤ + (V(0) − ) exp (−αt) ∀t ≥ 0. (7)
loop semi-explicit
 system (4) in the form of an ellipsoid α α
EP := {z ∈ R2  z T P z ≤ 1}, where P is a symmetric pos- Note that the presented Lemma 1 can be proved directly
itive defined n × n-dimensional matrix. Using the above- using the well-known properties of the linear systems and
mentioned classic definition of an asymptotically stable the Comparison Principle (see [17]). A function V from
set, we now introduce the useful ”attractive ellipsoid” Lemma 1 is further called a ”storage function”.
concept. Our aim is to show that under some additional as-
sumptions the combined function V (x(t)), where x(·) is a
Definition 5: We call EP an attractive ellipsoid for sys-
solution to (2) for a suitable gain matrix K ∈ Υ, satisfies
tem (4) if it is an asymptotically stable invariant set of
all the conditions of the Lemma 1. With this aim let
(4).
us use the Lyapunov-like analytic tool and introduce the
The chosen gain matrix K for the linear feedback causes following ”energetic” function V : Rn → R+ associated
a dynamic behavior of the complete state vector z(t) such with the original implicit system (2)
that the following inequality z T (t)P z(t) ≤ 1 holds. The
V (x) := xT P x, P = P T , P > 0.
last can be considered in an exact or any approximative
sense. The resulting dynamic behavior of the closed-loop Note that the symmetric positive defined matrix P here is
system can be interpreted as a stability in a ”practical” the same matrix as in (6). Let us consider this energetic
sense. function namely, V (x(t)) evaluated along the trajectories

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 33
978-1-4673-2169-3/12/$31.00 ©2012 IEEE
 2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control
Mexico City, Mexico. September 26-28, 2012

of the initial system (2). We use the additional notation Using Theorem 2, we now are able to characterize
σ(η(t), x(t)) := (η(t) + f (x(t)) − Ax(t)). We are now ready the abstract constraint set Υ in the main optimization
to formulate our main theoretic result. problem (6) by a constructive way. Recall that Υ ⊂ Rm×n
is the set of the admissible gain matrices that guarantee
Theorem 2: Let α > 0, β := 1 + δ > 0. Then the Lie
the attractiveness of the ellipsoid EP (see Definition 5).
derivative
Therefore, the main problem (6) has now the following
DV (x(t)) :=∇V (x(t))[Ax(t)+ form
(f (x(t)) − Ax(t) + BKx(t) + η(t))]
minimize tr{P −1 }
of the combined function V (x(·)) satisfies the following subject to P > 0, P = P T , (8)
inequality
W (P, P1 , K, α) < 0.
DV (x(t)) + αV (x(t)) − β ≤ Z T (t)W (P, P1 , K, α)Z(t),
with a symmetric block-matrix W (P, P1 , K, α). Let us now
where Z(t) = (xT (t), ẋT (t), σ T (η(t), x(t)))T is the ex- note that due to the bilinear structure of W (P, P1 , K, α)
tended state vector, P1 ∈ Rn×n and W (P, P1 , K, α) is a the negative definiteness of this matrix-function consti-
matrix-function. If W (P, P1 , K, α) is chosen as a negative tutes a numerically sophisticated question. The nonlinear
defined matrix, then the combined function V (x(·)) has nature of the obtained multidimensional inequality in (8)
all the properties of a storage function V(·). evidently involves a possible relaxation approach. We next
Proof: Let V (x) = xT P x and P1 ∈ Rn×n . Using the replace the above bilinear matrix inequality by an auxil-
idea of the ”descriptor method” (see [7] for details), we iary system of linear matrix inequalities (see [3] for LMIs).
can represent DV (x(t)) as follows This expected relaxed system will provide a basis for
the further numerical treatment of the main optimization
DV (x(t)) = DV (x(t))+
problem (8). Introduce the following symmetric matrix-
2h[P x(t) + P ẋ(t) + P1 σ(η(t), x(t))] , function
(−E ẋ(t) + (A + BK)x(t) + σ(η(t), x(t)))i,
Wr (P, P1 , Y, α) := M T W (P, P1 , K, α)M,
where the second addend in the last formula is an
effective zero. Applying this representation and taken where M := diag(P −1 , P −1 , P1−1 ) ∈ R3n×3n is an
into consideration the next useful effective zero given auxiliary matrix. The component wise definition of
by σ(η(t), x(t))T Qη σ(η(t), x(t)), we get DV (x(t)) + Wr (P, P1 , Y, α) can be given by the corresponding expres-
αV (x(t)) − β = Z T (t)W̃ (P, P1 , K, α)Z(t) − β + sions:
σ(η(t), x(t))T Qη σ(η(t), x(t)). Here we defined Wr11 = AP −1 + P −1 AT + BY + Y T B T + (α + h)P −1 ,
Wr12 = P −1 − EP −1 + P −1 AT + Y T B T ,
 
W̃11 W̃12 W̃13
W̃ :=  W̃21 W̃22 W̃23  Wr13 = P1−1 + P −1 AT + Y T B T ,
W̃31 W̃32 W̃33 Wr21 = P −1 − P −1 E T + AP −1 + BY,
Wr22 = −(EP −1 + P −1 E T ), Wr23 = P1−1 − P −1 E T ,
with Wr31 = P1−1 + AP −1 + BY,
W̃11 := P A + AT P + P BK + K T B T P + αP Wr32 = P1 − EP −1 , Wr33 = P1−1 − P1−1 Qη P1−1 ,
−1

W̃12 := P − P E + AT P + K T B T P
W̃13 := P + AT P1 + K T B T P1 where Y := KP −1 ∈ Rm×n is an artificial variable. The
W̃21 := P − E T P + P A + P BK last expressions for the component of Wr (P, P1 , K, α) are
linear functions (excepting the term Wr33 ) with respect to
W̃22 := −(P E + E T P ), W̃23 := P − E T P1
the basic variables (P, P1 , Y, ) for a fixed scalar component
W̃31 := P + P1T A + P1T BK
α. To avoid the technical problems with the bilinear
W̃32 := P − P1T E, W̃33 := P1T − Qη .
structure of Wr33 of the matrix Wr (P, P1 , Y, α) it is always
Using the boundedness conditions from Section II, we be possible to ask for a P1−1 which satisfies inequality
easily obtain kf (x) − Axk2Qη ≤ δ + hkxk2 , kηk2Qη ≤ 1.
We now put β = 1 + δ and derive the final estimation P1−1 Qη P1−1 > ρP1−1

DV (x(t)) + αV (x(t)) − β ≤ Z T (t)W (P, P1 , K, α)Z(t) for some real constant ρ > 0 and then majorize
Wr (P, P1 , K, α) < Fr (P, P1 , Y, α), where Fr (P, P1 , Y, α) is
where W (P, P1 , K, α) is the same matrix-function
a matrix that is equal to Wr (P, P1 , Y, α) with replacing
W̃ (P, P1 , K, α) replacing W̃11 by
of the element Wr33 by the over estimated term Fr33 :=
W11 := P A + AT P + P BK + K T B T P + (α + h)P. (1 − ρ)P1−1 . We can apply the celebrated Schur technique
to inequality above and to obtain
Evidently, the negative definiteness of W (P, P1 , K, α)  
guarantees the properties of the storage function V(·) (see Qη In
χ := > 0,
Lemma 1) for the combined function V (x(·)).  In ρ−1 P1−1

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 34
978-1-4673-2169-3/12/$31.00 ©2012 IEEE
 2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control
Mexico City, Mexico. September 26-28, 2012

where In is the identity matrix of the size n × n and ρ > 0 3) Get a solution α(l) ) to (10). Increase l = l + 1 and
is a suitable constant. Finally, in position of the original solve (9). If |P (l) − P (l−1) | < ǫ, go to Step (4). Else,
bilinear inequality in (8) we obtain the following matrix go to Step (2).
inequality that has a linear structure under assumption of 4) Stop. Determine (P opt , P1opt , Y opt , αopt ) :=
(l) (l) (l) (l) opt (l) (l)
fixed scalar parameter α: (P , P1 , Y , α )) and K := Y P .


W := diag Fr (P, P1 , Y, α), −χ < 0, The minimization problem (9) in Steps (1) and (3)
where χ is defined above. Note that all terms in W are constitutes an important part of the proposed algorithm.
linear. Therefore, the main optimization problem (6) has It can be solved using some standard numerical techniques.
now the following relaxed form The same is also true with respect to the auxiliary sub-
problem (10). Note that (9) (under assumption of the fixed
minimize tr{P −1 } scalar parameter α) is a linear optimization problem with
subject to P > 0, P = P T , (9) respect to the unknown matrices. Let us illustrate the
W(P, P1 , Y, α) < 0. effectiveness of the proposed Conceptual Algorithm I by
way of an academic example.
with a symmetric block-matrix W(P, P1 , Y, α).
Example 1: Consider the semi-explicit DAE system of
Theorem 3: Assume that the relaxed optimization prob- the type (1) with index one and concrete dates
lem (9) has an optimal solution (P opt , P1opt , Y opt ) for  
a fixed scalar variable α. Then the following 5-tuple 1 0

E= , f (x) = −0.1sign(x1 ), x2 ,
(P opt , P1opt , K opt , α) is also an optimal solution to the 0 0   
initial nonlinearly constrained problem (8) with K opt := −0.1 0 1
A= , B= .
Y opt P opt . 0 1 1
The last result is a direct consequence of the one-to- We evidently have the following simple estimations that
one transformations (introduced above) K = Y P for some corresponds to our main assumptions (see Section 2)
admissible P and Y . Therefore, the formal relaxation of kf (x) − Axk2 =| −0.1(sign(x1 ) − x1 ) |2 ≤ (0.1)2 + (0.1)2 x21
the initial problem does not really extend the solutions set
of the original problem (8). The obtained equivalent opti- We now fix h = δ = 0.01 and the initial value α(0) = 10−5 .
mization problem (9) with linear type restrictions (LMIs) Using the Conceptual Algorithm discussed above, we next
provides an analytic basis for the implementable numerical obtain  
opt 0.05879 0.00645
algorithms associated with the given problem (6) initially P = ,
0.00645 0.01235
formulated in the abstract form. Recall that (6) determines opt
K = (−0.056 − 0.032)
the practically stable linear feedback control strategy for
the implicit dynamic system (2). This control design is Using the obtained value K opt , and applying the standard
based on the minimal size attractive ellipsoid EP and in MATLAB toolboxes Ode23t, Ode15s and Ode15i, we are
fact defined by the selection of the optimal gain matrix able to calculate the trajectory x(·) of the closed-loop
K opt . system (2). We have considered the following initial condi-
T
IV. Numerical Aspects of the Attractive tions x0 = (20, 1.157) , ẋ0 = (−1.257, 10)T . The external
Ellipsoid Method perturbation is selected as η(t) = 1 × 10−3 (sin t, sin t)T .
The application of Conceptual Algorithm I in the given
In this part of the manuscript we apply the theoretic system induces the following numerical results indicated
results from Section III to the practical stability analysis on Fig. 2 and Fig. 3. Fig. 2 represents the state coordinates
of implicit systems (2). Our aim is to propose a realizable of the closed-loop system (2). The calculated optimal
attractive ellipsoid based numerical scheme for the com- matrix P opt determines the corresponding attractive el-
putational robust control design. Let us first summarize lipsoid of the minimal size that contains the origin. Fig.
the constructive approaches discussed above in the form 3 shows the time evolution of the same state components
of a Conceptual Algorithm. (x1 (t), x2 (t)).
Conceptual Algorithm 1: Finally note that the applied theoretic method and the
0) Put l = 0, select an initial positive scalar α(0) and corresponding computational results constitutes a robust
choose an accuracy ǫ > 0. behavior of the resulting system in the sense of the effective
1) Solve the optimization problem (9). disturbances rejection.
(l)
2) Use the obtained solution Φ(l) := (P (l) , P1 , Y (l) ) to
(9) and solve the system of linear inequalities (with V. Concluding Remarks
respect to the unknown pair α) In this paper we developed a new analytic and com-
(
(l)
putational method for the robust feedback control design
W(P (l) , P1 , Y (l) , α) < 0 of nonlinear implicit dynamic systems governed by semi-
(10)
α > 0. explicit DAEs of index one and closed by a linear control

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 35
978-1-4673-2169-3/12/$31.00 ©2012 IEEE
 2012 9th International Conference on Electrical Engineering, Computing Science and Automatic Control
Mexico City, Mexico. September 26-28, 2012

1.6
Phase−plane
[5] L. Dai, Singular Control Systems, Springer, Berlin, 1989.
1.4
[6] G.J. Duncan and F.C. Schweppe, Control of linear dynamic
PHASE PLANE
systems with set constrained disturbances, IEEE Transactions
1.2
on Automatic Control, AC-16, 1971, pp. 411 – 423.
1 [7] E. Fridman, Descriptor discretized Lyapunov functional
method: analysis and design, IEEE Transaction on Automatic
STATE X2(t)

0.8
Control, 51, 2006, pp. 890 – 897.
0.6
[8] W. Haddad and V. Chellaboina, Nonlinear Dynamical Systems
0.4
and Control, Princeton University Press, Princeton, 2008.
[9] H.K. Khalil, Nonlinear Systems, Prentice Hall, Upper Saddle
River, 2002.
0.2

0 [10] P. Kunkel and V. L. Mehrmann, Differential-Algebraic

−0.2
Equations: Analysis and Numerical Solutions, EMS Publishing
House, Switzerland, 2006.
−0.4
−5 0 5
STATE X1(t)
10 15 20
[11] A.B. Kurzhanski and P. Varaiya, Ellipsoidal techniques for
reachability under state constraints, SIAM Journal on Control
and Optimization, 45, 2006, pp. 1369 – 1394.
Fig. 2: Phase plane of nonlinear system. [12] M. Mera, A. Poznyak and V. Azhmyakov, On the robust control
design for a class of continuous-time dynamical systems with
a sample-data output, in: Proceedings of the 18th IFAC World
Congress, Milano, Ialy, 2011, pp. 5819 – 5824.
x1(t)

20
x2(t)
[13] A. N. Michel, L. Hou and D. Liu, Stability of Dynamical
Systems, Birkhäuser, New York, 2007.
[14] S. A. Nazyn, B. T. Polyak and M. V. Tpopunov, Rejection
AND X2(t)

15 TEMPORAL EVOLUTION OF THE STATE COMPONENTS of bounded exogenous disturbances by the method of invariant
ellipsoids, Automation and Remote Control, 3, 2007, pp. 467
X1(t)

– 486.
[15] B.T. Polyak and M.V. Topunov, Suppression of bounded exo-
STATE COMPONENTS

10

geneous disturbances: output control, Automation and Remote

Control, 69, 2008, pp. 801 – 818.
5
[16] A. Polyakov and A. Poznyak, Lyapunov function design
for finite-time convergence analysis: ”Twisting” controller for
second-order sliding mode realization, Automatica, 45, 2009,
0
pp. 444 – 448.
0 10 20
TIME (t)
30 40 50 [17] A. Poznyak, Advanced Mathematical Tools for Automatic
Control Engineers: Deterministic Techniques, Elsevier, Am-
sterdam, 2008.
Fig. 3: Temporal evolution of trajectories x1 and x2 . [18] A. Poznyak, V. Azhmyakov and M. Mera, Practical output
feedback stabilisation for a class of continuous-time dynamic
systems under sample-data outputs, International Journal of
Control, 84, 2011, pp. 1408 – 1416.
law. The proposed design procedure in our contribution [19] W. C. Rheinbolt, On the existence and uniqueness of so-
lutions of nonlinear semi-implicit differential-algebraic equa-
constitutes an extension of the conventional invariant tions. Nonlinear analysis, Theory, Methods and Applications.
ellipsoid technique (called the attractive ellipsoid method). 16, 1981, pp. 642 – 661
From the analytic point of view, the elaborated methodol- [20] K. Takaba, Stability analysis of interconnected implicit sys-
tems based on passivity, in: Proceedings of the 15th Triennial
ogy generates an admissible linear feedback control strat- World Congress, Barcelona, Spain, 2002.
egy that guarantee existence and a concrete characteriza- [21] E.D. Yakubovich, Solution of the optimal control problem for
tion of a minimal size invariant ellipsoid associated with the linear discrete systems, Automation and Remote Control,
36, 1976, pp. 1447 – 1453.
the system realization. The obtained ellipsoidal invariant
set is asymptotically stable in the sense of Lyapunov.
Finally note that the main analytic idea and the associated
numerical treatment can be easily applied to some alterna-
tive linear and nonlinear control processes that incorporate
some natural bounded uncertainties.

References
[1] V. Azhmyakov, On the geometric aspects of the invariant
ellipsoid method: application to the robust control design, in:
Proceedings of the 50th Conference on Decision and Control,
Orlando, USA, 2011, pp. 1353-1356.
[2] F. Blanchini and S. Miani, Set-Theoretic Methods in Control,
Birkhäuser, Boston, 2008.
[3] S. Boyd, E. Ghaoui, E. Feron, V. Balakrishnan, Linear
Matrix Inequalities in System and Control Theory, SIAM,
Philadelphia, 1994.
[4] M.A. Dahleh, J.B. Pearson and J. Boyd, Optimal rejection of
persistent disturbances, robust stability, and mixed sensitivity
minimization, IEEE Transactions on Automatic Control, 33,
1988, pp. 722 – 731.

IEEE Catalog Number: CFP12827-CDR

ISBN: 978-1-4673-2168-6 36
978-1-4673-2169-3/12/$31.00 ©2012 IEEE