Professional Documents
Culture Documents
CHP - 10.1007 - 978 3 642 24064 5 - 6
CHP - 10.1007 - 978 3 642 24064 5 - 6
CHP - 10.1007 - 978 3 642 24064 5 - 6
6.1 Introduction
In Section 2.6, the design of full-order and reduced-order state observers to estimate
the entire state vector of a class of Lipschitz nonlinear systems was discussed. This
chapter considers nonlinear systems where the nonlinear functions are not necessar-
ily to be Lipschitz. The focus is on the design of observers that can estimate a partial
set of the state vector and/or the unknown input vector for these nonlinear systems.
In Section 6.2 of this chapter, one considers a class of nonlinear systems and
discusses a design approach that enables one to design reduced-order observers to
estimate only a required partial set of the state vector. A key feature of this design
approach is the decomposition of the nonlinearities into two portions: one portion
is Lipschitz with respect to the partial states; the other portion comprises the re-
maining terms of the nonlinearities and is treated as unknown inputs. Based on this
decomposition, necessary and sufficient conditions are derived for the existence of
reduced-order observers. Then, conditions are derived for the solvability of the de-
sign matrices of the proposed observer and for the stability of its dynamics.
In Section 6.3 of this chapter, one extends the design method of Section 6.2 to
deal with a class of nonlinear systems which includes an unknown input vector.
Conditions for the existence and stability of reduced-order unknown input func-
tional observers are derived. To ensure asymptotic convergence of the error system,
a linear matrix inequality (LMI) is derived based on the Lyapunov function coupled
with the Lipschitz assumption of the nonlinear function.
In Section 6.4 of this chapter, one considers the design of full-order asymptotic
observers to estimate simultaneously both the state and unknown input vectors of
a class of multiple-input multiple-output nonlinear systems. Again, the nonlinear
function is not necessarily to be Lipschitz, rather it is treated more generally as a
combination of a state-dependent unknown function including additive disturbance
as well as uncertain/time-varying terms, and a nonlinear function which is Lips-
chitz with respect to the state and input. Upon the satisfaction of some existence
conditions, the observer design problem can be efficiently solved via a LMI-based
technique with asymptotic estimation guaranteed.
H. Trinh and T. Fernando: Functional Observers for Dynamical Systems, LNCIS 420, pp. 127–152.
springerlink.com
c Springer-Verlag Berlin Heidelberg 2012
128 6 Functional State and Unknown Input Observers for Nonlinear Systems
where x(t) ∈ Rn , u(t) ∈ Rm and y(t) ∈ R p are the state, input and the output vectors,
respectively. The nonlinear vector function f (x(t), u(t)) ∈ Rn is not required to be
Lipschitz. The vector z(t) ∈ Rr (r ≤ (n − p)) is to be estimated. Matrices A, B, C and
F are real constant and of appropriate dimensions. It is assumed that the pair (A, B)
is controllable and also
without
loss of generality, it is assumed that rank(C) = p,
C
rank(F) = r and rank = (p + r) ≤ n.
F
The problem considers in this section is the design of a reduced-order observer
to estimate the partial state vector z(t). To address this problem, one considers the
following reduced-order observer for the nonlinear system (6.1)-(6.3)
where w(t) ∈ Rq is the observer state vector which has a low order, where
r≤ q ≤ (n − p). Matrices D = [Ir 0], E, N, J, L and the nonlinear function
y(t)
f1 , u(t) ∈ Rn are to be determined such that ẑ(t) converges asymptoti-
ẑ(t)
cally to z(t).
As outlined in the Introduction, a key feature of the solution method in this chap-
ter lies in the decomposition of the nonlinear function f (x(t), u(t)) into two por-
tions: one portion is Lipschitz with respect to z(t) and the other portion comprises
the remaining terms of the nonlinearities and is treated as unknown inputs. The next
section will show how such a decomposition can be achieved.
and
q (t)
f (x(t), u(t)) = f V T 1 , u(t) . (6.9)
q2 (t)
Let the right-hand side of (6.9) be decomposed as follows
q (t) q1 (t)
f VT 1 , u(t) = f1 (ξ (t), u(t)) + f˜ , u(t) (6.10)
q2 (t) q2 (t)
where −1
T S ξ (t)
f1 (ξ (t), u(t)) = f V , 0 + f (0, u) (6.11)
0
and
˜f (x(t), u(t)) = f V T q1 (t) , u(t) − f1 (ξ (t), u(t))
q2 (t)
= f (x(t), u(t)) − f1 (Γ x(t), u(t)). (6.12)
where f2 (x(t), u(t)) ∈ Rl is now treated as an unknown input (or disturbance) vec-
tor and 0 ≤ l ≤ n is the number of independent unknown inputs. Matrix W ∈ Rn×l
is a full-column rank matrix. There is no loss of generality in assuming that ma-
trix W is of full-column rank. Otherwise, the following rank decomposition can be
applied to the matrix W : W f2 (x(t), u(t)) = W̃ W̄ f2 (x(t), u(t)), where W̃ is now a
full-column rank matrix and f¯2 (x(t), u(t)) = W̄ f2 (x(t), u(t)) can be considered as a
new unknown input vector.
The aforementioned development completes the decomposition of the nonlin-
ear function f (x(t), u(t)) into the form (6.6). Thanks to this decomposition (6.6),
one can now combine both the results of unknown input functional observers and
Lipschitz nonlinear systems to address the functional observer design problem for
nonlinear systems (6.1)-(6.3).
130 6 Functional State and Unknown Input Observers for Nonlinear Systems
Let L ∈ Rq×n be a full-row rank matrix and define the error vectors ε (t) ∈ Rq and
e(t) ∈ Rr to be
The following Theorem presents necessary and sufficient conditions for the exis-
tence of the observer (6.4)-(6.5).
Theorem 6.1.
The estimate ẑ(t) as defined in (6.4) is an asymptotic estimate of z(t) for the decom-
position of the nonlinearity as in (6.6) for all x(0), ẑ(0), u(t) and all possible set of
nonlinear functions f (x(t), u(t)) ∈ Rn if and only if the following conditions hold.
Condition 1: The error ε (t) determined by the following observer error system
y(t) y(t)
ε̇ (t) = N ε (t) + L f1 , u(t) − f1 , u(t) (6.16)
Dε (t) + z(t) z(t)
Remark 1: Observe that the matrix equations in the above Condition 2 are the same
as those matrix equations presented in Proposition 4.3 for linear functional observers
with unknown inputs. In addition to the normal requirement that the matrix N be
Hurwitz, one also needs the asymptotic convergence of the error system (6.16).
The proof of Theorem 6.1 can be constructed by following along similar lines of
proofs for the existence of linear functional unknown input observers and that of
state observers for Lipschitz nonlinear systems. For completeness, its proof is shown
next.
Proof (Sufficiency): From (6.14) and (6.1)-(6.6), the following error dynamics equa-
tion is obtained
Substituting (6.20) and (6.22) into the observer error system (6.16) given in
Condition 1 of Theorem 6.1 gives
where
1 1
Θ = Q(N1 −ZN2 )+(N1 −ZN2 )T Q+ QL̄1 L̄T1 Q+ QZ L̄2 L̄T2 Z T Q+ γ 2 (β1 + β2 )Iq .
β1 β2
(6.30)
From (6.29)-(6.30), by using the Schur decomposition and by letting G = QZ,
the following LMI can be obtained.
6.2 Partial State Estimation for Nonlinear Systems 133
Theorem 6.2.
The error ε (t), as defined in (6.16), converges asymptotically to zero provided that
the Lipschitz assumption (6.26) holds and there exist a positive definite matrix Q =
QT ∈ Rq×q , a matrix G ∈ Rq×p and two positive scalars β1 and β2 such that the
following LMI holds ⎡ ⎤
Δ QL̄1 GL̄2
⎣ L̄T1 Q −β1 In 0 ⎦<0 (6.31)
T
L̄2 G T 0 −β2 In
where Δ = QN1 + N1T Q − GN2 − N2T GT + γ 2 (β1 + β2 )Iq . Furthermore, matrix Z is
given by Z = Q−1 G.
Remark 3: For the case when matrix Γ is square, i.e. when the number of states
to be estimated reaches r = (n − p), then the decomposition (6.6) gives f1 (ξ , u) =
f (x, u) = f (Γ −1 ξ , u), f2 (x, u) = 0 and W = 0. If the nonlinearity f (x, u) is Lips-
chitz as assumed in [118, 146], then f1 (ξ , u) is also Lipschitz and vice-versa. The
procedure in this section can be used to design an (n − p) order observer. Further-
more, if the nonlinearity f (x, u) is non-Lipschitz, then the procedure in this section
allows the design of an (n − p) order observer if the decomposition (6.6) produces
a Lipschitz function f1 (ξ , u).
Let one now design a reduced-order observer to estimate a linear function, z(t),
where z(t) = Fx(t) = x4 (t) + x5 (t). First, the nonlinear function f (x, u) is decom-
posed according to (6.6) so that
T
f1 (ξ , u) = 0 0 0 0.45sin2 (ξ4 ) 0 0 0
T
W = 4 0 0.1 0 0 0 1
f2 (x, u) = x4 x7 , where f1 (ξ , u) is Lipschitz with a Lipschitz constant γ = 0.45 and
the nonlinearity function f2 (x, u) = x4 x7 is treated as unknown input.
First, one uses the transformation matrix P as defined in (2.21) and the following
submatrices are obtained
134 6 Functional State and Unknown Input Observers for Nonlinear Systems
⎡ ⎤
⎡ ⎤ ⎡ ⎤ 0 0 0
−1 0 0 1 0 0 0 ⎢ 0 0 0⎥
A11 = ⎣ 2 0 1 ⎦, A12 = ⎣ 0 1 0 0 ⎦, A21 = ⎢ ⎥
⎣ 1 0 0 ⎦, F1 = 0 0 0 ,
−2 3 0 −1 1 1 0
−1 1 0
⎡ ⎤ ⎡ ⎤
−3 0 1 1 ⎡ ⎤ 0
⎢ 0 1 0 −1 ⎥ 4 ⎢0⎥
A22 = ⎢⎣ 0 0 −1 0 ⎦ W1 =
⎥ ⎣ 4 ⎦, W2 = ⎢ ⎥
⎣ 0 ⎦, F2 = 1 1 0 0 , P1 = C
−3.9
0 1 0 −2 1
and P2 = 04×3 I4 .
Here, one starts with the lowest order, i.e., q = r = 1. For this, one first checks
the necessary condition (5.41). Here, the condition is satisfied and thus one next
checks the conditions (5.46)-(5.47) of Theorem 5.3 to see if they are satisfied or not.
If both are satisfied, then the solvability of all the matrix equations in Condition 2
of Theorem
⎡ 6.1 is achieved. Using⎤ (5.42)-(5.45) one obtained the following
−0.7071 0 0 4
Ω = ⎣ 0.7071 0 0 4 ⎦, Φ = −2.8284 −1 0 0 , N1 = −1.4875 and N2 =
1.4142 1 0 −3.9
0.
With the above matrices, one can easily check that both conditions (5.46)-(5.47)
of Theorem
5.3 are satisfied. Using (6.23)-(6.24),
L̄1 and L̄2 are obtained as follows
L̄1 = 0.0250 −1.4875 −1 1 1 0 0 and L̄2 = 0.
With the Lipschitz constant γ = 0.45, one can check that the LMI problem (6.31) is
feasible with the following parameters
β1 = β2 = 1, Q = 0.2796 and G = 0
L = 0.0250 −1.4875 −1 1 1 0 0 , E = −0.5125 1.4875 1
and
y(t) T
f1 = , u = 0 0 0 0.45sin2 (ẑ) 0 0 0 .
ẑ(t)
Fig. 6.1 shows the simulated responses of z(t) and ẑ(t), which shows that
ẑ(t) → z(t).
This section extends the results of Section 6.2 to include an external unknown in-
put vector. Consider the nonlinear system described by (6.1)-(6.3) with an added
unknown input vector, d(t), i.e.,
6.3 Partial State Estimation for Nonlinear Systems with Unknown Inputs 135
−2
−4
−6
−8 Actual Trajectory
Estimated Trajectory
−10
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8
Time
where d(t) ∈ Rq is the unknown input vector, D ∈ Rn×q and without loss of gen-
erality, it is assumed that matrix D has full-column rank. The nonlinear function
f (x, u) is decomposed according to (6.6) (i.e., f (x, u) = f1 (ξ , u) + W f2 (x, u)), as
shown in Section 6.2.1. It is assumed that the nonlinear function f1 (ξ , u) satisfies
the Lipschitz assumption defined in (6.26).
To estimate z(t), consider the following reduced-order observer of order r
Let L ∈ Rr×n be a full-row rank matrix and define the error vectors ε (t) ∈ Rr and
e(t) ∈ Rr to be
By following similar lines as in Section 6.2.2, the following necessary and suf-
ficient conditions for the existence of the reduced-order observer (6.35)-(6.36) can
be derived.
Theorem 6.3.
The estimate ẑ(t) as defined in (6.35)-(6.36) is an asymptotic estimate of z(t) for the
decomposition of the nonlinearity as in (6.6) for all x(0), ẑ(0), u(t), all possible set
of nonlinear functions f (x(t), u(t)) ∈ Rn and unknown inputs, d(t), if and only if the
following conditions hold.
Condition 1: The error ε (t) determined by the following observer error system
y(t) y(t)
ε̇ (t) = N ε (t) + L f1 , u(t) − f1 , u(t) (6.39)
ε (t) + z(t) z(t)
Observe that in the above condition 2, an extra matrix equation (i.e., LD = 0) arose
due to the addition of an unknown input vector d(t). Here, the design of a reduced-
order observer to estimate z(t) is again reduced to finding the matrices N, L, J and
E such that the two conditions in Theorem 6.3 are satisfied. One can adopts an
approach presented in Chapter 5 to derive these unknown matrices. On the other
hand, the following discussion also provides an alternative way to derive them.
The last matrix equation of condition 2 can be expressed as L = F − EC and by
substituting this into the first matrix equation of condition 2, one obtains
CA
NF = FA − E T (6.41)
C
where T = J − NE.
Post-multiplying both sides of (6.41) by the following non-singular matrix
+
F F⊥ (6.42)
Ir 0 0 0
Ω
= r + rank . (6.53)
ψ
138 6 Functional State and Unknown Input Observers for Nonlinear Systems
As can be seen from above, it is clear that (6.50) is equivalent to (6.51). This
completes the proof of Theorem 6.4.
Observe that the condition (6.50) can be easily tested as both sides involve all
known matrices. It is also clear that the condition (6.50) is a generalization of
the necessary condition for the existence of an unknown input observer. When
the
C
number of linear functions to be estimated reaches (n − p) (i.e., rank = n),
F
condition (6.50) is reduced to the well-known observer matching condition, i.e.,
rank(CD̄) = rank(D̄), where D̄ = [D W ].
Now, a general solution to (6.47) is given as
E T = ψΩ + + Z(I2p − Ω Ω + ) (6.55)
N = N1 − ZN2 (6.56)
L = L1 − ZL2 (6.59)
Thus, upon the satisfaction of the LMI (6.70), matrix gain Z can be derived from
Z = Q−1 G. Hence from (6.55), (6.56) and (6.59), matrices E, T , N and L can be
obtained. Then from T = J − NE, matrix J is obtained as J = T + NE. As as result,
all of the observer parameters are obtained.
Since N2 = 0 and L2 = 0, one can let Z = 0 and with γ = 3, one can easily check
that the LMI (6.70) is satisfied with β1 = 1. The rest of the observer parameters can
be obtained, where
−22 1 −2 0 1 0 2 0 1
N = N1 = , L = L1 = ,E= , LB = ,
−19.5 −6 −0.5 0 0 1 0.5 0 3.5
−23.5 −2 T
J= and f1 (ξ̂ (t)) = 0 0 0 3cos(ẑ1 (t)) .
−35.05 −0.5
142 6 Functional State and Unknown Input Observers for Nonlinear Systems
For numerical simulation, the control input signal u(t) and unknown input signal,
d(t), are shown in Fig. 6.2. Figures 6.3 and 6.4 show the responses of the system
states x3 (t) and x4 (t) and their estimates. It is clear from the simulation that all the
estimates settle quickly to the actual responses of these signals.
40
u(t)
d(t)
30
20
10
−10
−20
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
In the literature, there has been a considerable research focus on the problem of de-
signing asymptotic observers to estimate both the state and the unknown input vec-
tors of nonlinear systems (see, for example, [10, 14, 16, 46, 127, 145]). The problem
is motivated in part by applications requiring fault detection and diagnosis [13, 40],
or where measurement of the system inputs is either too expensive or practically
not possible [16]. In machine tool and manipulator applications, there are situations
where an input observer is required to estimate the cutting force of a machine tool
or the exerting force/torque of a robotic system. In chaotic systems, one wishes to
estimate not only the state for chaos synchronization but also the information signal
input for secure communication [10, 14, 75]. Input observers can enhance the per-
formance of automotive engine management systems [111–113] and improve the
bandwidth of micro-gyroscopes [31]. Unknown input observers have also found ap-
plication in various engineering areas such as wastewater treatment processes [57],
industrial biological processes [98] and fuel cell stack systems [8].
6.4 State and Input Simultaneous Estimation for Nonlinear Systems 143
10
Actual
Estimated
8
−2
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
15
10
−5
Actual
Estimated
−10
−15
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
This section discusses the problem of estimating simultaneously the state and un-
known input vectors of a class of nonlinear systems. The nonlinear function is not
necessarily required to be Lipschitz, rather it is treated more generally as a com-
bination of a state-dependent unknown function including additive disturbance as
well as uncertain/time-varying terms, and a nonlinear function which is Lipschitz
with respect to the state and input.
Consider a class of nonlinear systems as described by the following equations
[46]
where as usual, x(t) ∈ Rn and y(t) ∈ R p denote, respectively, the state vector and
known measured output vector. Here, u(t) ∈ Rm denotes the unknown input vec-
tor that one wishes to estimate and f (x, u, y) is a real nonlinear vector function on
Rn . Matrices A, B, C and D are real constant and of appropriate dimensions. The
nonlinear vector function f (x, u, y) considered in this section is not necessary to be
Lipschitz. Instead, by following along the lines of work as in Section 6.2, one can
decompose f (x, u, y) into two portions as follows
where f1 (x, u, y) ∈ Rn is now assumed to be Lipschitz in its first two arguments [10]
with a Lipschitz constant γ , i.e.,
where ζ̂ (t) denotes the state estimation vector of ζ (t) and w(t) ∈ R(n+m) . Matri-
ces E, N, J and L are to be determined such that ζ̂ (t) converges asymptotically to
ζ (t). It is clear from (6.75)-(6.76) that the proposed observer uses only the avail-
able output measurement y(t) and the known Lipschitz nonlinearity function to es-
timate the unknown vector ζ (t). Since ζ (t) contains both the unknowns x(t) and
u(t), if ζ̂ (t) → ζ (t), it follows that ζ̂ (t) provides both x̂(t) → x(t) and û(t) → u(t)
simultaneously.
and therefore the state and input estimation problem of (6.71)-(6.72) is now the
problem of designing an observer for the generalized (descriptor) system (6.78)-
(6.79) such that ζ̂ (t) converges asymptotically to ζ (t). This is essentially a state
observer design problem for a special class of nonlinear descriptor systems. The
following Theorem provides sufficient conditions for the existence of observers
(6.75)-(6.76).
Theorem 6.7.
For observer (6.75)-(6.76), the estimate ζ̂ (t) will converge asymptotically to ζ (t)
provided that the following conditions hold.
Condition 1: ⎧
⎪
⎪ LĒ + EC̄ = In+m
⎨
LW = 0
(6.80)
⎪
⎪ N = LĀ − GC̄
⎩
G = J − NE.
Condition 2: The error e(t) determined by the following observer error system
Proof: Define e(t) = ζ̂ (t) − ζ (t) as the error between ζ (t) and its estimate ζ̂ (t).
Using (6.75) and (6.79) the error e(t) can be expressed as
Therefore, provided that Condition 2 of Theorem 6.7 is also met, then ζ̂ (t) is an
asymptotic estimate of ζ (t). This completes the proof of Theorem 6.7.
From Theorem 6.7, the design of the observer is reduced to the problem of finding
matrices L, N, J and E so that Conditions 1 and 2 are satisfied.
The following Theorem presents a necessary and sufficient condition for the solv-
ability of all the matrix equations in Condition 1 of Theorem 6.7.
Theorem 6.8.
All the matrix equations
in Condition 1 of Theorem 6.7 are solvable if and only if
matrix D CW has full-column rank, i.e.,
rank D CW = (m + l). (6.88)
Remark 4: A close examination of the condition (6.88) reveals that the condition
is a generalization of the well-known observer matching condition. For condition
(6.88) to hold, a necessary condition is that the number of outputs is at least equal to
the number of unknown inputs (i.e., u(t) ∈ Rm ) plus the size of the nonlinear terms
which don’t satisfy the Lipschitz condition (i.e., f2 (x, u, y) ∈ Rl ). These two signals
affecting the plant dynamics are the only two signals in (6.71)-(6.72) which do not
satisfy any regularity condition: they are arbitrary and are treated as unknowns. On
top of that, it is required that the effects of these two signals are distinguishable in
some sense at the plant output, because of the full-column rank condition.
Proof: Equations LĒ + EC̄ = In+m and LW = 0 of Condition 1 can now be combined
into
L E S = In+m 0 (6.89)
6.4 State and Input Simultaneous Estimation for Nonlinear Systems 147
With W having a full-column rank of l, the left-hand side of (6.91) can be deter-
mined as
⎡ ⎤
Ē W
S
rank = rank ⎣ C̄ 0 ⎦
In+m 0
In+m 0
W
= (n + m) + rank
0
= (n + m + l). (6.92)
L = L1 + ZL2 , (6.95)
E = E1 + ZE2 (6.96)
148 6 Functional State and Unknown Input Observers for Nonlinear Systems
Using (6.95), matrix N defined in Condition 1 of Theorem 6.7 can now be ex-
pressed as
N = LĀ − GC̄ = Φ + ZΨ − GC̄ (6.99)
where Φ and Ψ are known matrices, defined as
In (6.99), Z and G are two unknown matrices which will be solved for while ensuring
that the observer error system defined in (6.101) will be asymptotically stable. Then
from G = J − NE, matrix J is obtained as J = G + NE. This completes the proof of
Theorem 6.8.
Let one now focuses on the asymptotic stability of the observer error defined
in Condition 2 of Theorem 6.7. By substituting (6.99) and (6.95) into (6.81), one
obtains
Under the Lipschitz assumption (6.74), the problem of finding Z and G to ensure
that the estimation error, e(t), converges asymptotically to zero for all ζ (t) and y(t)
can be quite easily handled. This problem has been discussed previously (see, for
example, from (6.25) and thereon by using a Lyapunov functional approach). Thus,
by following the same approach, the following theorem, expressed in LMI, is given
which ensures asymptotic convergence of the observer error.
Theorem 6.9.
The error e(t), as defined in (6.101), converges asymptotically to zero for all
ζ (t) and y(t) provided that the Lipschitz assumption (6.74) holds and there ex-
ist a positive definite matrix Q = QT ∈ R(n+m)×(n+m) , matrices X ∈ R(n+m)×p and
Y ∈ R(n+m)×(n+p) and two positive scalars β1 and β2 such that the following LMI
holds ⎡ ⎤
Δ QL1 Y L2
⎣ LT1 Q −β1 In 0 ⎦<0 (6.102)
LT2 Y T 0 −β2In
where Δ = QΦ + Φ T Q + Y Ψ + Ψ T Y T − C̄T X T − X C̄ + γ 2 (β1 + β2 )I(n+m) .
Furthermore, matrices Z and G are given by Z = Q−1Y and G = Q−1 X.
A computational algorithm is stated as follows.
6.4 State and Input Simultaneous Estimation for Nonlinear Systems 149
and the Lipschitz constant is γ = 3.205. The nonlinear term, tx2 (t)x3 (t), can now be
treated as unknown input.
Based on the design algorithm given in Section 6.4.2, the following step by step
is followed.
150 6 Functional State and Unknown Input Observers for Nonlinear Systems
Step 1: Using (6.97) and (6.100) matrices L1 , L2 , Φ and Ψ are obtained, where
⎡ ⎤ ⎡ ⎤
1 0 0 0 −10 1 0 0 1
⎢ 2 0 0 0⎥ ⎢ −20 2 0 0 2⎥
⎢ ⎥ ⎢ ⎥
L1 = ⎢
⎢ 0 0 1 0 ⎥, L2 = 0, Φ = ⎢ 0 0 −22 1 2 ⎥ and Ψ = 0.
⎥ ⎢ ⎥
⎣ 0 0 0 1⎦ ⎣ 1.95 0 −19.5 −6 0.5 ⎦
−2 0 0 0 20 −2 0 0 −2
Note that here L2 = 0 and Ψ = 0 and thus one can let Z = 0 and hence Y = 0.
Accordingly, the LMI problem (6.102) is reduced to only finding a positive matrix
Q, a matrix X and a positive scalar β1 .
Step 2: With Z = 0 and Y = 0, the LMI problem (6.102) yields the following
⎡ ⎤ ⎡ ⎤
2.0731 0.1205 0.0509 0.2659 0.2543 −10.7345 3.3533
⎢ 0.1205 1.7112 −0.6647 0.5416 0.3853 ⎥ ⎢ −23.0995 12.1190 ⎥
⎢ ⎥ ⎢ ⎥
Q= ⎢⎢ 0.0509 −0.6647 0.8233 −0.6269 −0.3101 ⎥, X=⎢ −2.2594 0.6919 ⎥
⎥ ⎢ ⎥
⎣ 0.2659 0.5416 −0.6269 0.8703 0.4124 ⎦ ⎣ −3.9436 −1.0486 ⎦
0.2543 0.3853 −0.3101 0.4124 0.9842 12.6836 −1.9851
and β1 = 1.
Steps 3&4: All the observer parameters of the observer defined in (6.75)-(6.76)
are obtained as follow
⎡ ⎤ ⎡ ⎤
−6.6195 −0.5009 0 0 1.1894 1 0 0 0
⎢ 2.7132 −9.0790 0 0 2.2776 ⎥ ⎢ 2 0 0 0⎥
⎢ ⎥ ⎢ ⎥
⎢
N = ⎢ 27.5118 −6.8983 −22 1 8.8575 ⎥, L = L1 = ⎢ ⎥
⎢ 0 0 1 0 ⎥,
⎥
⎣ 21.6775 1.7764 −19.5 −6 12.1402 ⎦ ⎣ 0 0 0 1⎦
−2.2514
⎡ 1.8242
⎤ ⎡0 0 −9.3015
⎤ −2 0 0 0
0 0 0 1
⎢ −2 1 ⎥ ⎢0 2⎥
⎢ ⎥ ⎢ ⎥
E = E1 = ⎢ 0 0 ⎥ and J = ⎢
⎢ ⎥ ⎥
⎢ 4 0 ⎥. Also, the term L f1 (ζ̂ (t), y(t)) can be
⎣ 0 0⎦ ⎣1 0⎦
2 0 0 −2
expressed as ⎡ ⎤ ⎡ ⎤
1 0 0 0 ⎡ ⎤ 0
⎢ 2 0 0 0⎥ 0 ⎢ ⎥
⎢ ⎥⎢ ⎥ ⎢ 0 ⎥
⎢ ⎥ ⎢ 0 ⎥ ⎢ ⎥.
L f1 (ζ̂ (t), y(t)) = ⎢ 0 0 1 0 ⎥ ⎣ ⎦ = ⎢ 0 ⎥
⎣ 0 0 0 1⎦ 0 ⎣ 3.205sin(x̂3 (t)) ⎦
3.205sin(x̂3 (t))
−2 0 0 0 0
For numerical simulation, the unknown input signal, u(t), is chosen as a sinu-
soidal signal, u(t) = 100sin9t. Superimposed in Fig. 6.5 are the time responses of
the input signal u(t) and its estimate, û(t). Figures 6.6 and 6.7 show the responses of
the system states x3 (t) and x4 (t) and their estimates. It is clear from the simulation
that all the estimates of both the input and system states settle quickly to the actual
responses of these signals.
6.4 State and Input Simultaneous Estimation for Nonlinear Systems 151
150
Actual
Estimated
100
50
−50
−100
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
25
Actual
Estimated
20
15
10
−5
−10
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
30
Actual
25 Estimated
20
15
10
−5
−10
−15
0 0.5 1 1.5 2 2.5 3 3.5 4
Time(sec)
6.5 Conclusion
This chapter has presented a method for the design of reduced-order observers to
estimate a required partial set of states of the state vector of a class of nonlinear
systems. A procedure has been presented for the decomposition of the nonlinear
function into two portions, one is Lipschitz with respect to the partial states and the
other is treated as unknown inputs. Based on this decomposition, observer design
can be realized by combining unknown input observers theory with nonlinear Lips-
chitz results. The results have also been extended to include an additional unknown
input vector. This chapter has also discussed the design of full-order observers to
estimate simultaneously the state and unknown input vectors for a class of nonlin-
ear multi-input multi-output systems. All design methods presented in this chapter
dealt with a wider class of nonlinear systems where the nonlinear functions are not
necessarily to be Lipschitz. In the next chapter, one will further look at the problem
of estimating simultaneously the state and unknown input vectors of uncertain sys-
tems which are subjected to time-varying delay. The emphasis will be on the design
of reduced-order observers and linear functional observers.