Anda di halaman 1dari 27

STRUCTURAL CONTROL AND HEALTH MONITORING

Struct. Control Health Monit. 2004; 11:161187 (DOI: 10.1002/stc.34)


H
1
energy control and its stability analysis for civil
engineering structures
Wen-Hwa Wu*
,y
and Chia-Chin Lin
Department of Construction Engineering, National Yunlin University of Science and Technology, Yunlin 640, Taiwan
SUMMARY
In order to assign clear physical meanings to H
1
control in its applications to civil structures under seismic
loads, an energy control methodology is established in this study. With delicate denition of the regulated
output vector, this H
1
control can be interpreted as limiting the ratio of the sum of system energy and
control energy with respect to the external excitation energy below a constant level. This limited value of
output-to-input ratio is usually denoted in the H
1
control literature by a conventional control parameter g:
Two dierent energy control formulations result from choosing either the strain or kinetic energy of a
structure to represent the system energy with an additional energy weighting parameter. Based on the
stability criterion of excluding pure imaginary or zero eigenvalues, the characteristic polynomials of the
Hamiltonian matrices corresponding to both energy control laws are also explored in this paper to further
derive analytical expressions for the greatest lower bound of g that guarantees the system stability. In
addition, from comparing the results for a demonstrative example, the control eectiveness of these energy
control algorithms is comprehensively investigated to induce several guidelines for selecting the control
parameters. Copyright # 2004 John Wiley & Sons, Ltd.
KEY WORDS: H
1
energy control; strain energy control; kinetic energy control; stability analysis;
Hamiltonian matrix
1. INTRODUCTION
The possibility of adopting active control technologies to reduce the vibration of civil
engineering structures was initially proposed in the 1970s [1]. Since then, the area of active
structural control has attracted intensive research eorts with considerable progress in
theoretical analysis [2] as well as laboratory simulation [36] and has now reached the stage
of practical application. The conventional linear quadratic regulator (LQR) or H
2
method was
popularly adopted in a large amount of this research because of its easy applicability. In
developing these optimal control algorithms, however, the external excitation has to be assumed
zero or in a pure white-noise distribution. This simplication may be appropriate for certain
systems under transient vibration or combined environmental noises, but is denitely
Received 12 June 2003
Published online 24 May 2004 Revised 1 September 2003
Accepted 8 September 2003 Copyright # 2004 John Wiley & Sons, Ltd.
y
E-mail: wuwh@yuntech.edu.tw
*Correspondence to: Wen-Hwa Wu, Department of Construction Engineering, National Yunlin University of Science
and Technology, Yunlin 640, Taiwan.
unreasonable for civil structures usually subjected to spectrum-specic wind or seismic forces.
Moreover, diculties such as model uncertainty, inability to measure the full state, time delay,
and actuator saturation are often encountered in practical cases of structural control, mostly
coming from the huge size and complexity of civil structures.
To tackle these problems, numerous advanced control theories have been developed for
applications in the control of structures under earthquake excitations. These methods include, for
instance, frequency domain optimal control [7], complete state (feedbackfeedforward) control
[8, 9], instantaneous optimal control [10], optimal direct output feedback control [1113], H
1
control [1418], etc. Among those new control laws, the H
1
optimal control is especially
promising since this approach accounts for all the possible excitation inputs and is consequently a
robust control methodology to the random environmental loads acted on civil structures.
Most of the early control theories, such as LQR control, are based on the one-way control
concept where the external disturbance or noise is ignored and the control force is solely
manipulated to deal with the structural response. H
1
control, on the other hand, is a two-way
control methodology where the control force and the external excitation are simultaneously
considered to constrain the innity norm of the transfer function between the excitation inputs
and a set of regulated outputs. This is in fact a two-person game in game theory with a goal of
exerting the least control force to withstand the worst external excitation. When rst introduced in
1981 [19], the concept of H
1
control did not immediately gain much attention because of its
corresponding mathematical diculties in analysis. However, the continuing work in the late
1980s [2023] successfully solved this problem by employing a unied framework of optimization.
Even though the analytical techniques for H
1
control have matured and provide a solid basis
for applications in various elds, many practical concerns in the eective control of civil
structures still require further exploration. For example, design engineers would usually ask how
the regulated outputs in H
1
control could be dened such that the structural control directly
corresponds to clear physical meanings and what values of the control parameters should be
adopted to attain the control goal at aordable cost. In this study, it is accordingly aimed to
establish an energy control methodology addressing the above issues. With delicate denition of
the regulated output vector, the H
1
control can be interpreted as limiting the ratio for the sum
of system energy and control energy with respect to the external excitation energy below a
constant level. This limited value of output-to-input ratio is usually denoted in the H
1
control
literature by a conventional control parameter g: Two dierent energy control formulations
result from choosing either the strain or kinetic energy of the structure to represent the system
energy with an additional energy weighting parameter. Based on these two energy control
formulations, stability analysis is further conducted in this paper to derive analytical expressions
for the greatest lower bound of g that guarantees the system stability. In addition, from comparing
the results for a demonstrative example, the eectiveness of these energy control algorithms is
comprehensively investigated to induce several guidelines for selecting the control parameters.
2. GENERAL H
1
CONTROL THEORY
2.1. State space description of an actively controlled linear structure system
Consider an actively controlled linear structure system represented by an n-DOF time-invariant
discrete-parameter model. If this system is subjected to a r 1 external excitation vector wt
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 162
and another l 1 control force vector ut is also applied, its equations of motion can then be
expressed as a second-order dierential equation:
%
MM
.
% xx% xxt
%
CC

% xx% xxt
%
KK% xxt
%
BBut
%
EEwt 0 1
where t denotes the time variable, % xxt represents the n 1 vector of structural displacements,
%
MM n n mass matrix of structure,
%
CC n n viscous damping matrix of structure,
%
KK n n
stiness matrix of structure,
%
BB n l location matrix of control forces and
%
EE n r location
matrix of external excitations. Equation (1) can be reformulated into state space and becomes a
rst-order dierential equation:
xxt Axt But Ewt 2
where xt h% xxt

% xx% xxti
T
is a 2n 1 state vector. In Equation (2), A, B, and E symbolize the
2n 2n system matrix, 2n l control input matrix and 2n r disturbance input matrix,
respectively, and take the form:
A
0
nn
I
nn

%
MM
1
%
KK
%
MM
1 %
CC
_ _
; B
0
nl

%
MM
1
%
BB
_ _
; E
0
nr

%
MM
1
%
EE
_ _
3
where 0 signies the zero matrix and I stands for the identity matrix, both with the
corresponding dimension indicated in the subscript. In addition, the p 1 output vector of the
system can be generally dened as a linear combination of the system state and the control force
vector:
zt Cxt Dut 4
where the p 2n matrix C and the p l matrix D are usually called the output matrix and the
direct transmission matrix, respectively.
2.2. H
1
norm and H
1
control
Various norms, usually denoted by jjvtjj; have been dened to measure a vector function vt:
The most commonly adopted norms in the control literature include the H
2
norm:
jjvtjj
2

_
1
1
jvtj
2
dt
_ _
1
2

_
1
1
v
T
tvt dt
_ _
1
2
5
and the H
1
norm:
jjvtjj
1
sup
t
jvtj 6
where j j represents the length of a vector and sup indicates that the least upper bound is taken.
As for measuring a system, it is most convenient to take the ratio of the output signal zt
relative to the input excitation signal wt: Consequently, the measurement of a system depends
on the norms adopted to measure the output and input signals. If both the output and input
signals of a system S are measured using H
2
norm, the H
1
norm of this system is dened as
jjSjj
1
sup
w
jjztjj
2
jjwtjj
2
7
Based on the denition in Equation (7), the primary concept of H
1
control is to design the
control force vector ut such that the output signal zt and the input excitation signal wt
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 163
always satisfy the following condition under any circumstances:
jjSjj
1
sup
w
jjztjj
2
jjwtjj
2
5g or
_
t
f
0
z
T
tzt dt5g
2
_
t
f
0
w
T
twt dt 8
where t
f
represents the terminal time of the control process. In Equation (8), g is a positive scalar
parameter, selected by the designer to guarantee that the ratio of output with respect to input is
constrained under this specic value. A smaller designated value of g means that more stringent
performance of the controlled system is required. For covering dierent types of output, the
formulation of Equation (8) is generally adopted in the H
1
control literature and the dimension
of g totally depends on the dimension of zt: For the energy control considered in this study, the
output term z
T
tzt in Equation (8) should be established with respect to a certain type of
energy, as will be comprehensively discussed in the next section. In this case, Equation (8) can be
deliberately redened as
jjSjj
1
sup
w
jjztjj
2
jjwtjj
2
5
g

K
1
_ or
_
t
f
0
z
T
tzt dt5
g
2
K
1
_
t
f
0
w
T
twt dt 9
where K
1
can be chosen as any stiness value, e.g., the eective stiness of the rst system
vibration mode. With the denition of Equation (9), the parameter g turns out to be
dimensionless and will simplify the presentation of many formulas in the following derivations,
especially in the section of stability analysis.
To mathematically analyze the H
1
control problem, a performance index is usually dened as
Ju; w
1
2
_
t
f
0
z
T
tzt
g
2
K
1
w
T
twt
_ _
dt 10
Since the excitation input wt tends to increase the performance index J; but the control force
input ut is designed to minimize J; Equation (9) can be further reformulated as
min
u
max
w
1
2
_
t
f
0
z
T
tzt
g
2
K
1
w
T
twt
_ _
dt min
u
max
w
J50 11
With the expression in Equation (11), the H
1
control is in fact a two-person game between the
external excitation wt and the control force ut: In other words, Equation (11) represents a
typical minmax problem where the worst situation is rst induced by the external excitation and
the control force is then introduced for improvement. Therefore, the H
1
control is usually
called the worst-case design.
2.3. H
1
state feedback control
If the type of state feedback is considered in designing the H
1
control law, the whole problem is
subsequently described as determining the control force in the form of
ut Gxt 12
such that Equation (11) can be satised under the state motion constraint given by Equation (2).
In Equation (12), G is conventionally called the feedback gain matrix. Calculus of variation has
been applied [22] to solve this constrained optimization problem of H
1
control and the resulting
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 164
solution takes the form of the algebraic Riccati equation (ARE) as follows:
PA BD
T
D
1
D
T
C A BD
T
D
1
D
T
C
T
P P
K
1
g
2
EE
T
BD
T
D
1
B
T
_ _
P
C
T
C C
T
DD
T
D
1
D
T
C 0 13
where P is the so-called Riccati matrix. The derivation of Equation (13) and the corresponding
optimal control force can be found in the literature [22] and will not be repeated in this paper. It
should be mentioned, however, that the ARE of Equation (13) diers slightly from that in the
literature, by replacing g
2
into g
2
=K
1
according to the modied denition for energy control as
shown in Equation (9). After solving for P, the optimal control force is obtained as
ut D
T
D
1
D
T
C D
T
D
1
B
T
P xt 14
In most practical applications, it is normally further assumed that
C
T
D 0
2nl
or D
T
C 0
l2n
15
With this condition, Equations (13) and (14) can be conveniently simplied to:
PA A
T
P P
K
1
g
2
EE
T
BD
T
D
1
B
T
_ _
P C
T
C 0 16
and
ut D
T
D
1
B
T
Pxt 17
In the conventional LQR optimal control [2], the performance index is typically dened as
J
1
2
_
t
f
0
x
T
tQxt u
T
tRut dt 18
where Q is the weighting matrix for the state vector and R denotes the weighting matrix for the
control force vector. The corresponding ARE of this problem is expressed as
PA A
T
P PBR
1
B
T
P Q 0 19
Comparing Equations (16) and (19), it is obvious that the H
1
control is totally equivalent to the
LQR optimal control when C
T
C Q; D
T
D R; and g ! 1:
3. H
1
ENERGY CONTROL
Various options, e.g. directly in terms of displacement or velocity, can be adopted to dene the
output vector zt for associating the H
1
control with apparent physical meaning in its
applications to structural systems. However, the choices with displacement or velocity will have
to assign D as a zero matrix and consequently make the control algorithm unfeasible because of
Equation (17). To avoid this problem, the aim is to relate the whole output term z
T
tzt in
Equation (10) to a certain type of energy and is accordingly called the H
1
energy control. To
attain this goal, the p 2n output matrix C and the p l direct transmission matrix D should be
deliberately specied such that Equation (15) can rst be satised to simplify Equation (4) as
z
T
tzt x
T
tC
T
Cxt u
T
tD
T
Dut 20
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 165
In Equation (20), the rst term on the right-hand side evidently corresponds to the system
energy and the second term directly relates to the control energy. As a result, z
T
tzt represents
the sum of these two energies and the H
1
control expressed by Equation (9) or (11) can be
subsequently interpreted as restricting this sum under a designated ratio of the energy produced
by the external excitation. Depending on the choice of dierent energies to characterize the
system energy, two types of energy control formulation will be developed in the following two
subsections.
3.1. Strain energy control
If it is intended to characterize the system energy with the elastic strain energy and enforce the
term u
T
D
T
Du in Equation (20) to hold the dimension of energy, C and D can be taken as
C

a
_
%
KK
1
2
0
nn
0
ln
0
ln
_
_
_
_
and D
1

K
1
_
0
nl
I
ll
_ _
21
where a is a nonnegative dimensionless weighting parameter. With the form of Equation (21), it
should be noted that Equation (15) is satised and Equation (20) can be reduced to
z
T
tzt a% xx
T
t
%
KK% xxt
1
K
1
u
T
tut 22
The term a% xx
T
%
KK% xx in Equation (22) is proportional to the elastic strain energy of the structural
system and the other term u
T
u=K
1
corresponds to the control energy. Therefore, when the
conventional control parameter g is specied in this case, it implies that the sum of the strain
energy and the control energy should be limited to a level lower than g with respect to the energy
produced by the external excitation w
T
w=K
1
: Under this constraint, the balance between the
strain energy of structure and the control energy is determined by the weighting parameter a:
More specically, increasing the value of a would lead to the requirement of larger control
energy and a smaller associated strain energy.
3.2. Kinetic energy control
On the other hand, if the kinetic energy is desired to represent the system energy, C and D can be
suggested to be of the form
C
0
ln
0
ln
0
nn

b
_
%
MM
1
2
_
_
_
_
and D
1

K
1
_
I
ll
0
nl
_ _
23
where b is another nonnegative dimensionless weighting parameter to regulate the kinetic
energy. Application of Equation (23) in Equation (20) provides
z
T
tzt b

% xx% xx
T
t
%
MM

% xx% xxt
1
K
1
u
T
tut 24
The term b

% xx% xx
T
%
MM

% xx% xx in Equation (24) is proportional to the kinetic energy of the structural system.
Thus, the manipulation among dierent energies in the case of kinetic energy control is similar
to that of strain energy control, except that the kinetic energy is adopted to replace the role of
the strain energy. The weighting parameter b is used to govern the allocation of the structural
kinetic energy and the control energy.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 166
3.3. Analysis algorithm
With the form of C and D in either Equation (21) or (23), the state feedback control law of
Equation (17) can be reduced to
ut K
1
B
T
Pxt 25
Substitution of Equation (25) into the system state Equation (2) results in
xxt A K
1
BB
T
Pxt Ewt A
1
xt Ewt 26
where A
1
represents the controlled system matrix. Equation (26) is a rst-order dierential
equation and can be solved as
xt e
A
1
t
x0
_
t
0
e
A
1
tt
Ewt dt 27
For direct applications to practical cases of structural control, Equations (25) and (27) need to
be further discretized as
ut
k1
K
1
B
T
Pxt
k1
; k 1; 0; 1; . . . 28
and
xt
k1
e
A
1
Dt
xt
k

Dt
2
e
A
1
Dt
2
Ewt
k
wt
k1
; k 0; 1; 2; . . . 29
where t
k
kDt; t
k1
t
k
Dt; and the one-point Gaussian integration rule is employed.
Before applying the H
1
energy control proposed in this study, the conventional control
parameter g and the additional energy weighting parameter a or b have to be selected by the
design engineer. With the given control parameters and available system parameters, the ARE
of Equation (16) can then be solved to obtain the Riccati matrix P: Following the above o-line
work, the on-line analysis algorithm for H
1
energy control can be summarized as:
1. Set the initial system state vector x0 and initial external excitation vector w0;
2. Use Equation (28) to determine the initial control force vector u0 by taking k 1;
3. Apply Equation (29) to calculate xt
1
by taking k 0;
4. Use the result from step 3 and apply Equation (28) to compute ut
1
by taking k 0;
5. Repeat steps 3 and 4 by taking k 1; 2; . . . to calculate all the corresponding quantities
xt
k1
and ut
k1
:
4. STABILITY ANALYSIS
Compared with LQR optimal control, the parameter g is added in H
1
control to constrain the
system output under this specic ratio of the system input. It has been shown in the literature
[24] that there exists a unique positive denite solution (Riccati matrix) for the ARE
corresponding to the conventional LQR control and the stability of the controlled system can be
consequently guaranteed. With the introduction of g; however, the positive deniteness of the
Riccati matrix and the control stability is no longer assured for H
1
control unless g is restricted
to be greater than a specic value. Except for certain special cases, there are no general
analytical formulas in the literature for determining this greatest lower bound (g.l.b.) of g and
numerical procedures such as the bi-section method have to be utilized. Based on the newly
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 167
dened energy control formulations, special eorts are attempted in this section to develop an
analytical methodology such that the g.l.b. of g for a general H
1
controlled system can be
eciently obtained.
4.1. Stability criterion
With the denitions of C and D either for the strain energy control shown in Equation (21) or
for the kinetic energy control shown in Equation (23), D
T
D I
ll
=K
1
and the ARE of Equation
(16) can be written as
PA A
T
P P K
1
1
g
2
EE
T
BB
T
_ _ _ _
P C
T
C 0 30
To solve this ARE, it is rst assumed that
A
2
A K
1
1
g
2
EE
T
BB
T
_ _ _ _
P 31
Besides, let X denote the diagonal matrix constituted by all the eigenvalues of A
2
and U
represent the matrix formed by all the corresponding eigenvectors. With the further assumption
P
%
PPU
1
32
and following the detailed derivation described in the literature [25], the original nonlinear ARE
of Equation (30) to solve for the 2n 2n Riccati matrix P can be transformed into a linear
eigenvalue problem
A K
1
1
g
2
EE
T
BB
T
_ _
C
T
C A
T
_

_
_

_
U
%
PP
_ _

U
%
PP
_ _
X 33
where
H
A K
1
1
g
2
EE
T
BB
T
_ _
C
T
C A
T
_

_
_

_ 34
is usually referred to as the Hamiltonian matrix. After this eigenvalue problem is solved, the
Riccati matrix P can be obtained from Equation (32).
It should be noted that H is a 4n 4n matrix with a total of 4n eigenvalues and 4n
corresponding eigenvectors. However, Equation (33) merely contains 4n 2n simultaneous
equations and the Riccati solution P for the ARE of Equation (30) is a 2n 2n matrix.
Therefore, only 2n eigenvalues and their 2n corresponding eigenvectors of H are needed for
Equation (32) in solving P: The subsequent question is then: which 2n eigenvalues and
eigenvectors should be adopted among 4n of them? It has been shown in the literature [25] that
the 2n eigenvalues with negative real parts have to be selected from all the eigenvalues of H in
the determination of P such that the system stability can be guaranteed. In addition, it has also
been proved that the eigenvalues of the Hamiltonian matrix H appear in pairs of l
1
; l
2
;
. . . ; l
2n
[26]. With this feature, at most 2n eigenvalues with negative real parts can be obtained
for H: A stability criterion for the H
1
controlled system can consequently be established by
prohibiting any eigenvalues of H locating on the imaginary axis of the complex plane. Based on
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 168
this stability criterion, it will be shown in the following subsections that the theoretical formula
for the g.l.b. of the control parameter g can be analytically established for a general single-
degree-of-freedom (SDOF) system with the application of either H
1
strain energy control or
H
1
kinetic energy control.
4.2. Strain energy control
For an SDOF system with mass
%
MM m; stiness
%
KK K
1
k; and damping
%
CC c; its natural
frequency and damping ratio can be expressd as o

k=m
_
and z c=2mo: If the H
1
strain
energy control is applied,
%
BB 1 and
%
EE 1 in this case and its 4 4 Hamiltonian matrix H is in
the form
H
0 1 0 0
o
2
2zo 0
o
2
m
1
g
2
1
_ _
amo
2
0 0 o
2
0 0 1 2zo
_

_
_

_
35
To solve the eigenvalue problem of H; its characteristic determinant should be expanded into
the corresponding fourth-order characteristic polynomial:
detH lI
44

det
l 1 0 0
o
2
l 2zo 0
o
2
m
1
g
2
1
_ _
amo
2
0 l o
2
0 0 1 2zo l
_

_
_

_
l
4
2o
2
1 2z
2
l
2
o
4
1 a
1
g
2
1
_ _ _ _
l
4
a
2
l
2
a
0
36
where
a
2
2o
2
1 2z
2
and a
0
o
4
1 a
1
g
2
1
_ _ _ _
37
It is noteworthy that this polynomial contains only the terms in even orders. Consequently, its
corresponding eigenvalues must appear in pairs of opposite sign, as indicated in the previous
subsection.
Solving the characteristic polynomial shown in Equation (36) directly leads to
l
2

a
2

a
2
2
4a
0
_
2
38
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 169
To satisfy the stability criterion that excludes pure imaginary or zero eigenvalues, l
2
is not
allowed to hold negative or zero real values. In other words, the two roots for l
2
illustrated in
Equation (38) can only be a pair of complex conjugates or two positive real numbers. For the
former case to be true, it is necessary that
a
2
2
4a
0
4o
4
4z
2
z
2
1 a
1
g
2
1
_ _ _ _
50 ) g
2
>
a
a 4z
2
1 z
2

39
On the other hand, for the validity of the latter case, it is required that
a
2
2
4a
0
4o
4
4z
2
z
2
1 a
1
g
2
1
_ _ _ _
50 ) g
2
4
a
a 4z
2
1 z
2

40
a
2
2o
2
1 2z
2
50 ) z4

2
_
2
41
and
a
0
o
4
1 a
1
g
2
1
_ _ _ _
> 0 ) g
2
>
a
a 1
42
Since
a
a 4z
2
1 z
2

5
a
a 1
the concurrent satisfaction of Equations (40), (41), and (42) leads to
z4

2
_
2
and
a
a 1
5g
2
4
a
a 4z
2
1 z
2

43
Finally, the complete stability criterion can be obtained by taking the union of these two cases
and combining Equations (39) and (43) as
g >

a
a 4z
2
1 z
2

_
g
g:l:b:
when z5

2
_
2
g >

a
a 1
_
g
g:l:b:
when z5

2
_
2
_

_
44
where g
g:l:b:
denotes the greatest lower bound of the H
1
control parameter g:
From Equation (44), it is apparent that g
g:l:b:
is always less than one. Moreover, g
g:l:b:
is merely
related to the strain energy weighting parameter a if the original system damping ratio z is no
less than

2
_
=2: When z is less than

2
_
=2; however, g
g:l:b:
turns out to depend on both a and z:
Figure 1(a) illustrates how the value of g
g:l:b:
varies with z for dierent values of a: It is rst
observed that g
g:l:b:
generally decreases with increasing value of z if z5

2
_
=2 and stays at a
constant level if z5

2
_
=2: Besides, g
g:l:b:
is also found to increase with increasing value of a: For
most of the civil structures whose damping ratios are much less than

2
_
=2 (usually less than
10%), it is also demonstrated in Figure 1(a) that the value of g
g:l:b:
is very close to one no matter
what value of a is adopted. Therefore, g
g:l:b:
1 can be used as a rule of thumb for designing the
parameter g if the H
1
strain energy control is applied in practical cases.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 170
4.3. Kinetic energy control
If H
1
kinetic energy control is applied to an SDOF system, the corresponding fourth-order
characteristic polynomial of its Hamiltonian matrix H can be expanded as:
detH lI
44

det
l 1 0 0
o
2
l 2zo 0
o
2
m
1
g
2
1
_ _
0 0 l o
2
0 bm 1 2zo l
_

_
_

_
l
4
o
2
21 2z
2
b
1
g
2
1
_ _ _ _
l
2
o
4
l
4
b
2
l
2
b
0
45
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
(a)
0 0.2 0.4 0.6 0.8 1
0
0.2
0.4
0.6
0.8
1
(b) Damping Ratio
1 . 0
1
10
=
=
=

1 . 0
1
10
=
=
=

g
.
l
.
b
.

g
.
l
.
b
.

Figure 1. Greatest lower bound of SDOF structural system under H


1
energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 171
where
b
2
o
2
21 2z
2
b
1
g
2
1
_ _ _ _
and b
0
o
4
> 0 46
The polynomial of Equation (45) can be rst solved as
l
2

b
2

b
2
2
4b
0
_
2
47
To ensure the system stability, the two roots for l
2
as shown in Equation (46) can be a pair of
complex conjugates only where
b
2
2
4b
0
o
4
21 2z
2
b
1
g
2
1
_ _ _ _
2
4
_ _
50 )
b 4z
2
b
>
1
g
2
>
b 4z
2
1
b
48
or two positive real numbers where
b
2
2
4b
0
50 )
1
g
2
5
b 4z
2
b
or
1
g
2
4
b 4z
2
1
b
49
and
b
2
21 2z
2
b
1
g
2
1
_ _ _ _
50 )
1
g
2
5
b 22z
2
1
b
50
Since
b 4z
2
b
>
b 22z
2
1
b
>
b 4z
2
1
b
51
the intersection of Equations (49) and (50) results in the criterion
1
g
2
4
b 4z
2
1
b
52
By taking the union of the two cases represented by Equations (48) and (52), the total stability
criterion can be obtained as
g >

b
b 4z
2

g
g:l:b:
53
Dierent from the case of strain energy control, the greatest lower bound of g in
kinetic energy control is governed by a single formula as shown in Equation (53).
Equation (53) also indicates that g
g:l:b:
is always less than one, and depends on the
kinetic energy weighting parameter b and the damping ratio z: Figure 1(b) depicts how
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 172
the value of g
g:l:b:
varies with z for several values of b: Similar to the results for strain energy
control as shown in Figure 1(a), g
g:l:b:
generally increases with increasing value of b
and decreases with increasing value of z: Also, for civil structures usually with light damping,
g
g:l:b:
1 is a very good approximation for designing the parameter g if H
1
kinetic energy
control is applied.
5. NUMERICAL EXAMPLE
A typical SDOF structural system is taken as an example in this section to examine the
eectiveness of the two proposed H
1
energy control algorithms and to investigate the guidelines
for selecting control parameters. The structural mass, stiness, and damping are assumed to be
m 345:6 ton; k 115:48 kN=m; and c 99:996 ton=s; respectively, which result in a natural
frequency of o 0:921 Hz and a damping ratio of z 2:5%: In addition, the ground
accelerations of the 1940 El Centro earthquake and the 1985 Mexico City earthquake are
considered as the external excitation.
5.1. Controlled natural frequency and damping ratio
For state feedback control, the control force is solely related to the system state and the
reduction in system vibration is accomplished totally by alternating the system matrix to
increase the stiness and damping of the original structure. Consequently, the control
performance can be directly assessed via examining the natural frequency and damping ratio of
the controlled system.
From Equation (26), the controlled system matrix A
1
for an SDOF system can be expressed in
terms of its corresponding 2 2 Riccati matrix P as
A
1
A K
1
BB
T
P
0 1
o
2
2zo
_ _
k
0
m
1
_ _
0 m
1

P
11
P
12
P
21
P
22
_ _

0 1
o
2
1
P
21
m
_ _
2o z
oP
22
2m
_ _
_

_
_

_
0 1
o
2
c
2z
c
o
c
_ _
54
where o
c
and z
c
are the natural frequency and damping ratio of the controlled SDOF system,
respectively. It is obvious from Equation (54) that only the elements in the second row of the
Riccati matrix P are involved in determining A
1
: With the ARE of Equation (30), all the four
elements of P for an SDOF system can be solved with the four corresponding quadratic
equations. Under the constraint that P must be symmetric and positive denite, a unique
solution of P can be analytically obtained. For the strain energy control, P
21
and P
22
take the
form
P
21
m
1

1 a
1
g
2
1
_ _

1
g
2
1
_ _
_

_
_

_
P
22

m
o
2z

4z
2
2 2

1 a
1
g
2
1
_ _

_
1
g
2
1
_ _
_

_
_

_
55
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 173
In the case of kinetic energy control,
P
21
0 P
22

m
o
2z

4z
2
b
1
g
2
1
_ _

1
g
2
1
_ _
_

_
_

_
56
It should be noted that the expressions in Equations (55) and (56) can still be valid for the case
of g 1 if LHospitals rule is applied to take the limit. Furthermore, examination of Equations
(55) and (56) reveals another convenient approach to establish the analytical formula for g
g:l:b:
:
Since the Riccati matrix has to be real, the quantity inside every square root sign in these two
equations is accordingly required to be positive. With this condition, the formulas of Equations
(44) and (53) can be easily obtained for the strain and kinetic energy control.
After P
21
and P
22
are solved, o
c
and z
c
of the controlled SDOF system can be conveniently
obtained. For strain energy control,
o
c
o

1
1

1 a
1
g
2
1
_ _

1
g
2
1
_ _

_
z
c
z

1
2

4
2
z
2
1

1 a
1
g
2
1
_ _
_ _

_
2
1
g
2
1
_ _

1
1

1 a
1
g
2
1
_ _

1
g
2
1
_ _

_
57
For kinetic energy control,
o
c
o
1
z
c
z
1
2

4
b
z
2
1
g
2
1
_ _

2
1
g
2
1
_ _ 58
From Equations (57) and (58), it is clear that the ratio of the controlled natural frequency o
c
to
the original natural frequency o is determined by the two parameters a and g in the case of
strain energy control, and o
c
is even identical to o for kinetic energy control. As to the ratio of
z
c
to z; it can be found in both cases that the relative change of damping ratio depends not only
on a (or b) and g; but also on z itself.
To investigate how the eectiveness of the strain energy control is inuenced by the control
parameter g; the normalized natural frequency and damping ratio, as expressed in Equation
(57), are plotted in Figure 2 for various values of a: It is found that the normalized damping
ratio increases with decreasing value of g when g
g:l:b:
5g42 and is extremely raised when g
approaches g
g:l:b:
: For larger values of g; however, z
c
=z is virtually reduced to a constant level
corresponding to the LQR control g ! 1: Moreover, the normalized frequency is basically
not aected by the value of g and the magnitude of o
c
=o is much less than z
c
=z; indicating that
the control eectiveness is primarily achieved by the alternation of system damping. For the case
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 174
of kinetic energy control, the controlled system frequency is even unchanged, and the variation
of z
c
=z with g is displayed in Figure 3 where similar trends are observed as in the case of strain
energy control. Therefore, it can be generally concluded that g should be kept below the value of
two in practical designs of H
1
energy control even though any value greater than g
g:l:b:
is
Figure 2. Normalized natural frequency and damping ratio of SDOF structural system
under H
1
strain energy control.
Figure 3. Normalized damping ratio of SDOF structural system under H
1
kinetic energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 175
theoretically feasible. Otherwise, the H
1
energy control would induce almost the same results as
the LQR control.
For further inspecting the control eectiveness in dierent ranges of the energy weighting
parameter a or b; the curves of o
c
=o and z
c
=z are depicted in Figure 4 for various values of g:
First of all, it is evident that o
c
=o (in the case of strain energy control) and z
c
=z increase
monotonically with increasing value of a or b: This trend is expected because the greater value of
energy weighting parameter leads to the smaller values of corresponding system energy as
previously discussed. For small values of a or b50:01; however, both o
c
=o and z
c
=z are
reduced to one, as in the uncontrolled case. Hence, values of a or b greater than 0.01 should
usually be adopted such that the control can make a dierence. It is also noteworthy in Figure 4
that the curves for these two types of H
1
energy control almost overlap when the energy
weighting parameter is less than the value of one, but a substantial dierence can be
distinguished for larger values of a and b: This trend further indicates that either strain or
kinetic energy control result in similar control eectiveness if the energy weighting parameter is
chosen to be less than one. Closer examination reveals that z
c
=z in the case of kinetic energy
control is always larger than that in the case of strain energy control if the same value larger
than one is taken for a and b:
Figure 4. Normalized natural frequency and damping ratio of SDOF structural system under
dierent H
1
energy control laws.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 176
5.2. Comparison of controlled responses and control force
In addition to checking the alternation of the controlled system parameters, the time histories of
the SDOF structural system subjected to El Centro earthquake and Mexico City earthquake are
directly examined in this subsection. Other than the reduction in structural responses, the
required control force is another crucial concern in evaluating the control eectiveness of a
control algorithm. The uncontrolled case and three dierent control cases of g ! 1LQR; g
1:33; and g ! g
g:l:b:
are compared. As to the energy weighting parameters, two values of 0.1 and
1 are selected for a or b: For the strain energy control cases adopting these parameter values,
their corresponding time-histories of structural displacement, structural acceleration,
and required control force are plotted in Figures 57 for the El Centro earthquake. For
Figure 5. Displacement of SDOF structural system subject to the 1940 El Centro earthquake
and under H
1
strain energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 177
clear comparison, only the portion of the rst 20 s in main earthquake excitation is displayed
in these gures. The maximum (MAX) and root-mean-square (RMS) values of the
interested quantities in these cases are also listed in Tables I and III for both
earthquakes. The time-histories for the kinetic energy control cases are displayed in Figures
810 for the Mexico City earthquake with the strong-motion portion of 1030 s: The
corresponding MAX and RMS values are also summarized in Tables II and IV for both
earthquakes.
From these gures and tables, it is generally observed that more signicant reduction in
dierent response categories can be achieved by selecting a smaller value of g at the expense of
providing larger control force. Examining various response quantities, it is also evident that the
Figure 6. Acceleration of SDOF structural system subject to the 1940 El Centro earthquake
and under H
1
strain energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 178
reduction in acceleration is not as much as that in displacement, and the decrease of the
maximum response is usually less than that of the RMS response. More specically, there exists
no considerable dierence between the results for g ! 1 and g 1:33; while much greater
suppression of responses can be attained in the case of g ! g
g:l:b:
with the requirement of larger
control forces. This tendency agrees with the trend illustrated in Figures 2 and 3 where z
c
=z is
sensitively increased only if g is very close to g
g:l:b:
: Comparing the control eectiveness for
dierent values of the energy weighting parameters, it is clear that all the response quantities are
decreased, and the control force is increased with increasing value of a or b: In addition, the
results are similar for both types of energy control if the energy weighting parameter takes the
value of 0.1 or 1. Again, all the above phenomena can be explained from the observed trends for
z
c
=z shown in Figure 4.
6. CONCLUSIONS
To give lucid physical meanings to the H
1
control in its applications to civil structures under
seismic loads, two types of energy control formulation including stain energy control and kinetic
Figure 7. Control force of SDOF structural system subject to the 1940 El Centro earthquake
and under H
1
strain energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 179
energy control are established in this study. In these two H
1
energy control algorithms, the
conventional parameter g is specied to limit the ratio for the sum of system energy and control
energy with respect to the external excitation energy below this constant level. Under this
constraint, the balance between the system energy of structure and the control energy is then
determined by the additional energy weighting parameter a or b: The performance of structure is
consequently governed by the parameters appearing in the above two dierent phases. In other
words, more signicant reduction in the structural response can be attained by either adopting a
smaller value of g or a larger value of the energy weighting parameter, at the expense of
requiring a larger control force. In the limiting case of g ! 1 where the eect of external
Figure 8. Displacement of SDOF structural system subject to the 1985 Mexico City earthquake
and under H
1
kinetic energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 180
excitation is not considered, H
1
control is reduced to conventional LQR control and only the
parameter a or b is solely in charge of deciding the structural performance.
Based on the stability criterion of excluding pure imaginary or zero eigenvalues, the
characteristic polynomials of the Hamiltonian matrices corresponding to both energy control
laws are also explored in this paper for a general SDOF structural system to further derive
analytical expressions for the greatest lower bound of g: From these formulas, it can be easily
concluded that g
g:l:b:
is always less than one, generally increases with increasing value of a or b;
and decrease with increasing value of z: For most of civil structures, usually with light damping,
it is also found that the value of g
g:l:b:
is very close to one and g
g:l:b:
1 can be used as a rule of
thumb for applying the energy control in practical cases.
Figure 9. Acceleration of SDOF structural system subject to the 1985 Mexico City earthquake
and under H
1
kinetic energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 181
In addition, by comparing the results for the example of an SDOF system, the control
eectiveness of the new energy control algorithms is comprehensively investigated and
compared to induce several guidelines for selecting the control parameters. First of all, values
of a or b must be greater than 0.01 such that the corresponding energy control can have distinct
dierence from the uncontrolled case. Secondly, g
g:l:b:
5g42 should be taken to ensure that H
1
energy control is not reduced to LQR control. Finally, diverse control eectiveness can be
obtained for the algorithms of strain energy and kinetic energy control only when the energy
weighting parameter is no less than 1.
For simplicity, the stability analysis and parametric study in this paper is focused on the
SDOF structural system even though the two H
1
energy control algorithms are generally
developed for any multi-degree-of-freedom (MDOF) system. Generalization of the analytical
methodology in stability analysis to determine the greatest lower bound of g for an MDOF
system may further require performing the procedures of reduction-of-order and diagonaliza-
tion to simplify the characteristic determinant of the higher-order Hamiltonian matrices.
Moreover, the system energy is represented by either the strain energy or the kinetic energy in
this work, which leads to two separate energy control laws. If the control formulation can be
reorganized such that the system energy may simultaneously include both the strain and kinetic
energies, a unied H
1
energy control algorithm will be subsequently established. Exploration of
the above extensions to this study is under way and will be reported in the near future.
Figure 10. Control force of SDOF structural system subject to the 1985 Mexico City earthquake
and under H
1
kinetic energy control.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 182
T
a
b
l
e
I
.
C
o
m
p
a
r
i
s
o
n
o
f
d
i

e
r
e
n
t
q
u
a
n
t
i
t
i
e
s
o
f
a
S
D
O
F
s
t
r
u
c
t
u
r
a
l
s
y
s
t
e
m
s
u
b
j
e
c
t
t
o
t
h
e
E
l
C
e
n
t
r
o
e
a
r
t
h
q
u
a
k
e
a
n
d
u
n
d
e
r
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l
.
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l

0
:
1

Q
u
a
n
t
i
t
y
N
o
c
o
n
t
r
o
l
g
!
1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
8
7
7
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
2
(
c
m
)
d
2

d
1
d
1
(
%
)
d
3
(
c
m
)
d
3

d
1
d
1
(
%
)
d
4
(
c
m
)
d
4

d
1
d
1
(
%
)
M
A
X
1
4
.
8
3
6
.
9
4

5
3
.
2
2
5
.
8
0

6
0
.
8
7
2
.
0
3

8
6
.
3
2
R
M
S
0
3
.
3
7
1
.
2
4

6
3
.
2
9
1
.
0
5

6
8
.
9
8
0
.
3
7

8
8
.
9
8
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
2
(
g
a
l
)
a
2

a
1
a
1
(
%
)
a
3
(
g
a
l
)
a
3

a
1
a
1
(
%
)
a
4
(
g
a
l
)
a
4

a
1
a
1
(
%
)
M
A
X
7
4
3
.
7
0
4
4
3
.
5
9

4
0
.
3
5
4
1
1
.
0
1

4
4
.
7
3
4
0
4
.
3
2

4
5
.
6
3
R
M
S
1
2
3
.
8
8
0
6
4
.
6
3

4
7
.
8
3
0
5
9
.
1
5

5
2
.
2
5
0
3
5
.
3
7

7
1
.
4
5
C
o
n
t
r
o
l
f
o
r
c
e
c
1
(
k
N
)
c
2
(
k
N
)
c
2

c
1
c
1
(
%
)
c
3
(
k
N
)
c
3

c
1
c
1
(
%
)
M
A
X
2
7
5
.
4
3
3
6
3
.
6
5
3
2
.
0
3
8
7
3
.
4
3
2
1
7
.
1
2
R
M
S
0
4
2
.
6
8
0
5
1
.
3
4
2
0
.
3
1
1
0
3
.
3
8
1
4
2
.
2
5
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l

1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
9
8
8
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
5
(
c
m
)
d
5

d
1
d
1
(
%
)
d
6
(
c
m
)
d
6

d
1
d
1
(
%
)
d
7
(
c
m
)
d
7

d
1
d
1
(
%
)
M
A
X
1
4
.
8
3
3
.
2
6

7
8
.
0
4
2
.
4
7

8
3
.
3
3
0
.
3
1

9
7
.
9
4
R
M
S
0
3
.
3
7
0
.
6
2

8
1
.
7
0
0
.
4
6

8
6
.
3
6
0
.
0
5

9
8
.
5
2
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
5
(
g
a
l
)
a
5

a
1
a
1
(
%
)
a
6
(
g
a
l
)
a
6

a
1
a
1
(
%
)
a
7
(
g
a
l
)
a
7

a
1
a
1
(
%
)
M
A
X
7
4
3
.
7
0
4
8
7
.
3
6

3
4
.
4
7
4
6
9
.
5
1

3
6
.
8
7
1
9
7
.
5
4

7
3
.
4
4
R
M
S
1
2
3
.
8
8
0
4
9
.
3
6

6
0
.
1
5
0
4
2
.
2
9

6
5
.
8
6
0
1
5
.
3
1

8
7
.
6
4
C
o
n
t
r
o
l
f
o
r
c
e
c
5
(
k
N
)
c
6
(
k
N
)
c
6

c
5
c
5
(
%
)
c
7
(
k
N
)
c
7

c
5
c
5
(
%
)
M
A
X
7
1
7
.
3
9
8
4
6
.
7
1
1
8
.
0
3
8
9
7
.
4
7
2
5
.
1
0
R
M
S
0
8
4
.
1
5
0
9
5
.
7
8
1
3
.
8
2
1
1
5
.
6
5
3
7
.
4
4
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 183
T
a
b
l
e
I
I
.
C
o
m
p
a
r
i
s
o
n
o
f
d
i

e
r
e
n
t
q
u
a
n
t
i
t
i
e
s
o
f
a
S
D
O
F
s
t
r
u
c
t
u
r
a
l
s
y
s
t
e
m
s
u
b
j
e
c
t
t
o
t
h
e
E
l
C
e
n
t
r
o
e
a
r
t
h
q
u
a
k
e
a
n
d
u
n
d
e
r
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l
.
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l

0
:
1

Q
u
a
n
t
i
t
y
N
o
c
o
n
t
r
o
l
g
!
1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
8
7
7
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
2
(
c
m
)
d
2

d
1
d
1
(
%
)
d
3
(
c
m
)
d
3

d
1
d
1
(
%
)
d
4
(
c
m
)
d
4

d
1
d
1
(
%
)
M
A
X
1
4
.
8
3
6
.
8
7

5
3
.
6
6
5
.
7
8

6
1
.
0
6
2
.
0
6

8
6
.
1
3
R
M
S
0
3
.
3
7
1
.
2
4

6
3
.
2
2
1
.
0
5

6
8
.
8
7
0
.
3
8

8
8
.
8
7
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
2
(
g
a
l
)
a
2

a
1
a
1
(
%
)
a
3
(
g
a
l
)
a
3

a
1
a
1
(
%
)
a
4
(
g
a
l
)
a
4

a
1
a
1
(
%
)
M
A
X
7
4
3
.
7
0
4
3
7
.
4
5

4
1
.
1
8
4
0
7
.
1
3

4
5
.
2
6
4
0
1
.
9
6

4
5
.
9
5
R
M
S
1
2
3
.
8
8
0
6
3
.
2
7

4
8
.
9
2
0
5
8
.
1
5

5
3
.
0
6
0
3
5
.
2
6

7
1
.
5
3
C
o
n
t
r
o
l
f
o
r
c
e
c
1
(
k
N
)
c
2
(
k
N
)
c
2

c
1
c
1
(
%
)
c
3
(
k
N
)
c
3

c
1
c
1
(
%
)
M
A
X
2
7
5
.
8
2
3
6
2
.
7
0
3
1
.
5
0
8
6
6
.
8
4
2
1
4
.
2
8
R
M
S
0
4
1
.
9
8
0
5
0
.
6
1
2
0
.
5
6
1
0
3
.
1
3
1
4
5
.
6
8
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l

1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
9
8
8
M
A
X
1
4
.
8
3
d
5
(
c
m
)
d
5

d
1
d
1
(
%
)
d
6
(
c
m
)
d
6

d
1
d
1
(
%
)
d
7
(
c
m
)
d
7

d
1
d
1
(
%
)
M
A
X
1
4
.
8
3
3
.
0
4

7
9
.
5
1
2
.
5
3

8
2
.
9
4
0
.
3
1

9
7
.
9
0
R
M
S
0
3
.
3
7
0
.
6
2

8
1
.
6
2
0
.
4
8

8
5
.
7
4
0
.
0
5

9
8
.
4
9
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
5
(
g
a
l
)
a
5

a
1
a
1
(
%
)
a
6
(
g
a
l
)
a
6

a
1
a
1
(
%
)
a
7
(
g
a
l
)
a
7

a
1
a
1
(
%
)
M
A
X
7
4
3
.
7
0
4
2
9
.
6
5

4
2
.
2
3
4
2
6
.
8
0

4
2
.
6
1
1
9
7
.
6
2

7
3
.
4
3
R
M
S
1
2
3
.
8
8
0
4
5
.
1
3

6
3
.
5
7
0
3
9
.
9
0

6
7
.
7
9
0
1
5
.
3
4

8
7
.
6
2
C
o
n
t
r
o
l
f
o
r
c
e
c
5
(
k
N
)
c
6
(
k
N
)
c
6

c
5
c
5
(
%
)
c
7
(
k
N
)
c
7

c
5
c
5
(
%
)
M
A
X
6
6
2
.
1
6
7
7
6
.
0
6
1
7
.
2
0
8
9
5
.
2
2
3
5
.
2
0
R
M
S
0
7
8
.
7
9
0
9
1
.
8
1
1
6
.
5
2
1
1
5
.
5
4
4
6
.
6
4
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 184
T
a
b
l
e
I
I
I
.
C
o
m
p
a
r
i
s
o
n
o
f
d
i

e
r
e
n
t
q
u
a
n
t
i
t
i
e
s
o
f
a
S
D
O
F
s
t
r
u
c
t
u
r
a
l
s
y
s
t
e
m
s
u
b
j
e
c
t
t
o
t
h
e
M
e
x
i
c
o
C
i
t
y
e
a
r
t
h
q
u
a
k
e
a
n
d
u
n
d
e
r
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l
.
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l

0
:
1

Q
u
a
n
t
i
t
y
N
o
c
o
n
t
r
o
l
g
!
1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
8
7
7
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
2
(
c
m
)
d
2

d
1
d
1
(
%
)
d
3
(
c
m
)
d
3

d
1
d
1
(
%
)
d
4
(
c
m
)
d
4

d
1
d
1
(
%
)
M
A
X
1
0
.
8
0
3
.
8
6

6
4
.
2
9
3
.
5
3

6
7
.
3
5
1
.
3
4

8
7
.
6
1
R
M
S
0
2
.
3
8
0
.
8
1

6
6
.
0
5
0
.
6
8

7
1
.
4
7
0
.
2
5

8
9
.
5
2
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
2
(
g
a
l
)
a
2

a
1
a
1
(
%
)
a
3
(
g
a
l
)
a
3

a
1
a
1
(
%
)
a
4
(
g
a
l
)
a
4

a
1
a
1
(
%
)
M
A
X
4
6
6
.
7
2
2
4
3
.
9
0

4
7
.
7
4
2
3
5
.
5
3

4
9
.
5
4
1
6
8
.
4
9

6
3
.
9
0
R
M
S
0
8
7
.
8
0
0
4
3
.
5
2

5
0
.
4
3
0
3
9
.
8
8

5
4
.
5
8
0
2
5
.
0
2

7
1
.
5
0
C
o
n
t
r
o
l
f
o
r
c
e
c
1
(
k
N
)
c
2
(
k
N
)
c
2

c
1
c
1
(
%
)
c
3
(
k
N
)
c
3

c
1
c
1
(
%
)
M
A
X
1
3
1
.
8
7
1
6
0
.
6
3
2
1
.
8
1
3
6
7
.
0
6
1
7
8
.
3
5
R
M
S
0
2
7
.
9
2
0
3
3
.
3
6
1
9
.
4
9
0
6
8
.
4
6
1
4
5
.
2
0
H
1
s
t
r
a
i
n
e
n
e
r
g
y
c
o
n
t
r
o
l

1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
9
8
8
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
5
(
c
m
)
d
5

d
1
d
1
(
%
)
d
6
(
c
m
)
d
6

d
1
d
1
(
%
)
d
7
(
c
m
)
d
7

d
1
d
1
(
%
)
M
A
X
1
0
.
8
0
2
.
1
0

8
0
.
5
2
1
.
5
0

8
6
.
1
2
0
.
1
7

9
8
.
4
4
R
M
S
0
2
.
3
8
0
.
4
0

8
2
.
9
6
0
.
3
0

8
7
.
2
3
0
.
0
3

9
8
.
5
3
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
5
(
g
a
l
)
a
5

a
1
a
1
(
%
)
a
6
(
g
a
l
)
a
6

a
1
a
1
(
%
)
a
7
(
g
a
l
)
a
7

a
1
a
1
(
%
)
M
A
X
4
6
6
.
7
2
2
0
8
.
7
5

5
5
.
2
7
1
8
4
.
1
7

6
0
.
5
4
5
5
.
1
4

8
8
.
1
9
R
M
S
0
8
7
.
8
0
0
3
3
.
8
0

6
1
.
5
0
0
2
9
.
3
6

6
6
.
5
6
0
7
.
2
0

9
1
.
8
0
C
o
n
t
r
o
l
f
o
r
c
e
c
5
(
k
N
)
c
6
(
k
N
)
c
6

c
5
c
5
(
%
)
c
7
(
k
N
)
c
7

c
5
c
5
(
%
)
M
A
X
2
7
8
.
3
8
3
3
8
.
7
4
2
1
.
6
8
5
3
8
.
2
2
9
3
.
3
4
R
M
S
0
5
5
.
1
7
0
6
2
.
9
2
1
4
.
0
4
0
9
9
.
7
5
8
0
.
8
1
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 185
T
a
b
l
e
I
V
.
C
o
m
p
a
r
i
s
o
n
o
f
d
i

e
r
e
n
t
q
u
a
n
t
i
t
i
e
s
o
f
a
S
D
O
F
s
t
r
u
c
t
u
r
a
l
s
y
s
t
e
m
s
u
b
j
e
c
t
t
o
t
h
e
M
e
x
i
c
o
C
i
t
y
e
a
r
t
h
q
u
a
k
e
a
n
d
u
n
d
e
r
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l
.
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l

0
:
1

Q
u
a
n
t
i
t
y
N
o
c
o
n
t
r
o
l
g
!
1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
8
7
7
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
2
(
c
m
)
d
2

d
1
d
1
(
%
)
d
3
(
c
m
)
d
3

d
1
d
1
(
%
)
d
4
(
c
m
)
d
4

d
1
d
1
(
%
)
M
A
X
1
0
.
8
0
3
.
9
8

6
3
.
1
3
3
.
6
0

6
6
.
6
7
1
.
3
5

8
7
.
4
6
R
M
S
0
2
.
3
8
0
.
8
0

6
6
.
3
4
0
.
6
8

7
1
.
4
8
0
.
2
5

8
9
.
4
1
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
2
(
g
a
l
)
a
2

a
1
a
1
(
%
)
a
3
(
g
a
l
)
a
3

a
1
a
1
(
%
)
a
4
(
g
a
l
)
a
4

a
1
a
1
(
%
)
M
A
X
4
6
6
.
7
2
2
4
0
.
0
2

4
8
.
5
7
2
3
2
.
8
2

5
0
.
1
2
1
6
8
.
1
5

6
3
.
9
7
R
M
S
0
8
7
.
8
0
0
4
2
.
5
2

5
1
.
5
7
0
3
9
.
2
0

5
5
.
3
6
0
2
4
.
9
6

7
1
.
5
8
C
o
n
t
r
o
l
f
o
r
c
e
c
1
(
k
N
)
c
2
(
k
N
)
c
2

c
1
c
1
(
%
)
c
3
(
k
N
)
c
3

c
1
c
1
(
%
)
M
A
X
1
3
2
.
8
0
1
6
1
.
1
0
2
1
.
3
1
3
6
5
.
4
3
1
7
5
.
1
7
R
M
S
0
2
7
.
2
4
0
3
2
.
7
8
2
0
.
3
0
0
6
8
.
3
0
1
5
0
.
7
1
H
1
k
i
n
e
t
i
c
e
n
e
r
g
y
c
o
n
t
r
o
l

1
g

1
:
3
3
g
!
g
g
:
l
:
b
:

0
:
9
9
8
8
D
i
s
p
l
a
c
e
m
e
n
t
d
1
(
c
m
)
d
5
(
c
m
)
d
5

d
1
d
1
(
%
)
d
6
(
c
m
)
d
6

d
1
d
1
(
%
)
d
7
(
c
m
)
d
7

d
1
d
1
(
%
)
M
A
X
1
0
.
8
0
2
.
1
5

8
0
.
1
1
1
.
6
8

8
4
.
4
5
0
.
1
7

9
8
.
4
0
R
M
S
0
2
.
3
8
0
.
4
1

8
2
.
7
9
0
.
3
2

8
6
.
5
2
0
.
0
4

9
8
.
4
6
A
c
c
e
l
e
r
a
t
i
o
n
a
1
(
g
a
l
)
a
5
(
g
a
l
)
a
5

a
1
a
1
(
%
)
a
6
(
g
a
l
)
a
6

a
1
a
1
(
%
)
a
7
(
g
a
l
)
a
7

a
1
a
1
(
%
)
M
A
X
4
6
6
.
7
2
1
9
9
.
8
3

5
7
.
1
8
1
8
0
.
0
9

6
1
.
4
1
5
5
.
1
6

8
8
.
1
8
R
M
S
0
8
7
.
8
0
0
3
1
.
0
9

6
4
.
5
9
0
2
7
.
8
5

6
8
.
2
8
0
7
.
2
0

9
1
.
8
0
C
o
n
t
r
o
l
f
o
r
c
e
c
5
(
k
N
)
c
6
(
k
N
)
c
6

c
5
c
5
(
%
)
c
7
(
k
N
)
c
7

c
5
c
5
(
%
)
M
A
X
2
5
7
.
0
9
3
1
1
.
7
9
2
1
.
2
8
5
3
7
.
9
3
1
0
9
.
2
3
R
M
S
0
5
1
.
5
4
0
6
0
.
3
8
1
7
.
1
4
0
9
9
.
7
0
0
9
3
.
4
3
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
W.-H. WU AND C.-C. LIN 186
ACKNOWLEDGEMENTS
The authors are grateful to the nancial support from the National Science Council of the Republic of
China under Grant NSC89-2211-E-224-033. In addition, the help of Mr Jian-Li Zeng in preparing several
gures and tables of this paper is also appreciated.
REFERENCES
1. Yao JTP. Concept of structural control. Journal of the Structural Engineering Division (ASCE) 1972; 98:15671574.
2. Soong TT. Active Structural Control: Theory and Practice. Longman: New York, 1990.
3. Chung LL, Reinhorn AM, Soong TT. Experiment on active control of seismic structures. Journal of the Engineering
Mechanics Division (ASCE) 1988; 114:241256.
4. Warnitchai P, Fujino Y, Benito MP, Agret R. An experimental study on active tendon control of cable-stayed
bridges. Earthquake Engineering and Structural Dynamics 1993; 22:93111.
5. Kobori T, Takahashi M, Nasu T, Niwa N. Seismic response controlled structure with active variable stiness
system. Earthquake Engineering and Structural Dynamics 1993; 22:925941.
6. Indrawan B, Kobori T, Sakamoto M, Ohrui S. Experimental verication of bound-force control method.
Earthquake Engineering and Structural Dynamics 1996; 25:179193.
7. Spencer BF, Suhardjo J, Sain MK. Frequency domain optimal control strategies for aseismic protection. Journal of
Engineering Mechanics Division (ASCE) 1994; 120:135157.
8. Suhardjo J, Spencer BF, Sain MK. Feedback-feedforward control of structures under seismic excitation. Structural
Safety 1990; 8:6989.
9. Hong HK, Liu CS, Liou DY. Complete state LQ optimal control of earthquake-excited structures. Proceedings of
the National Science Council, ROC (A) 1994; 18:386399.
10. Yang JN, Akbarpour A, Ghaemmaghami P. New optimal control algorithms for structural control. Journal of the
Engineering Mechanics Division (ASCE) 1987; 113:13691387.
11. Chung LL, Lin CC, Chu SY. Optimal direct output feedback of structural control. Journal of the Engineering
Mechanics Division (ASCE) 1993; 119:21572173.
12. Chung LL, Lin CC, Lu KH. Time-delay control of structures. Earthquake Engineering and Structural Dynamics
1995; 24:687701.
13. Lin CC, Lu KH, Chung LL. Optimal discrete-time structural control using direct output feedback. International
Journal of Engineering Structures 1996; 18:472480.
14. Schmitendorf WE, Jabbari F, Yang JN. Robust control techniques for buildings under earthquake excitations.
Earthquake Engineering and Structural Dynamics 1994; 23:539552.
15. Chase JG, Smith HA. Robust H
1
control considering actuator saturation I: theory. Journal of the Engineering
Mechanics Division (ASCE) 1996; 122:976983.
16. Chase JG, Smith HA, Suzuki T. Robust H
1
control considering actuator saturation II: applications. Journal of the
Engineering Mechanics Division (ASCE) 1996; 122:984993.
17. Kose IE, Schmitendorf WE, Yang JN. H
1
active seismic response control using static output feedback. Journal of
the Engineering Mechanics Division (ASCE) 1996; 122:651659.
18. Wu JC, Yang JN, Schmitendorf WE. Reduced-order H
1
and LQR control for wind-excited tall buildings.
International Journal of Bridge Engineering 1998; 20:222236.
19. Zames G. Feedback and optimal sensitivity: model reference transformations, multiplicative seminorms, and
approximate inverses. IEEE Transactions on Automatic Control 1981; 26:301320.
20. Petersen IR. Disturbance attenuation and H
1
optimization: a digital method based on the algebraic Riccati
equation. IEEE Transactions on Automatic Control 1987; 32:427429.
21. Glover K, Doyle JC. State-space formulae for all stabilizing controllers that satisfy an H
1
norm bound and relations
to risk sensitivity. Systems and Control Letters 1988; 11:167172.
22. Doyle JC, Glover K, Khargonekar PP, Francis BA. State-space solutions to standard H
2
and H
1
control problems.
IEEE Transactions on Automatic Control 1989; 34:831847.
23. Khargonekar PP, Petersen IR, Zhou K. Robust stabilization of uncertain linear systems: quadratic stabilizability
and H
1
control theory. IEEE Transactions on Automatic Control 1990; 35:356361.
24. Kwakernaak H, Sivan R. Linear Optimal Control Systems. Wiley: New York, 1972.
25. Meirovitch L. Dynamics and Control of Structures. Wiley: New York, 1990.
26. Potter B. Matrix quadratic solutions. SIAM Journal of Applied Mathematics 1966; 14:496501.
Copyright # 2004 John Wiley & Sons, Ltd. Struct. Control Health Monit. 2004; 11:161187
H
1
ENERGY CONTROL AND ITS STABILITY ANALYSIS 187

Anda mungkin juga menyukai