Documente Academic
Documente Profesional
Documente Cultură
In a previous paper [Baratta and Corbi, 1999] one has defined a procedure allowing to
identify a closed-loop control algorithm with feedback based on the whole record of the
response time-history rather than on instantaneous response parameters. The control
force results from control of each harmonic component of the forcing function, simply
integrated over the frequency domain. Every harmonic is controlled, independently of
each other, by a classical linear control whose coefficients are calibrated in way to make
the relevant response component a minimum compatibly with the control effort one
wants to apply at the corresponding frequency. The distribution of this control inten-
sity over the frequency range remains a arbitrary choice; such a choice however lends
itself to be effectively assisted by intuition, much more than similar choices in other
procedures (e.g.: the coefficients of the quadratic norms in the J-index optimization).
The result is that every harmonic remains controlled by a different couple of optimal
coefficients (corresponding to the proportional and to the derivative terms in the linear
control law), and the overall control force for an arbitrary disturbance, after Fourier
inverse transformation, is produced by feedback integration over the whole response
time-history.
The procedure, tested with reference to simple and composed harmonic excitations
incoming a s.d.o.f. structural system, has proved a good agreement of the numerical re-
sults with the theoretical treatment; furthermore it has shown that the main limit of such
an approach consists of referring the dynamic equilibrium solution to a particular solu-
tion, that, neglecting the initial conditions, may introduce some unstable components
in the oscillation. In the paper the effects induced in the controlled structural system
response by the adoption of the proposed procedure are deepened and an improved strat-
egy is presented, able to overcome the detrimental transient effects determined by the
original algorithm. The final adopted control law is shown to achieve an improved time
response, both in the transient and in the steady-state field, in comparison to a control
strategy based on classical linear control minimizing the response norm conditioned by
a bounded control.
1. Introduction
In the last years, control techniques are being increasingly applied to civil struc-
tures, with the purpose, in particular, to improve protection of buildings against
earthquakes and to improve safety and comfort against strong winds. The great
297
July 23, 2002 10:28 WSPC/124-JEE 00071
the search for the control parameters, optimal for an harmonic excitation, can be
set in the form of the following problem of constrained minimum
u2o (ω|p, q)
min = min H 2 (ω|p, q)
p,q fo2 (ω) p,q
(9)
2
w (ω|p, q)
sub = [p (ω) + ω q (ω)]H (ω|p, q) ≤ Co (ω)
2 2 2 2 2
fo2 (ω)
where Co (ω) is, for any given ω, a number less than unity, in that, as a rule, the
control force is generally required to be instantaneously smaller in magnitude than
the active force. With the position
with CIo = 1/Co . In order to solve the constrained extremum problem, let consider
the Kuhn–Tucker’s conditions, with λ the Lagrangian multiplier
∂G2 (ω|p, q) ∂V 2 (ω|p, q)
(1 + λ) − λC 2
(ω) =0
∂p Io
∂p
2 2
(1 + λ) ∂G (ω|p, q) − λC 2 (ω) ∂V (ω|p, q) = 0
Io
∂q ∂q (12)
λ[G (ω|p, q) − CIo (ω)V (ω|p, q)] = 0
2 2 2
λ≥0
yielding
(1 + λ)[k + p(ω) − mω 2 ] − λp(ω)CIo 2
(ω) = 0
(1 + λ)ω 2 [c + q(ω)] − λCIo
2
(ω)ω 2 q(ω) = 0
(13)
λ[G2 (ω|p, q) − CIo
2
(ω)V 2 (ω|p, q)] = 0
λ ≥ 0.
July 23, 2002 10:28 WSPC/124-JEE 00071
The case λ = 0 shall be discarded, in that it leads to the absolute minimum of the
objective function, i.e. G2 (ω|p, q) = 0. Assuming λ > 0 Eqs. (13) yield
(k − mω 2 )(1 + λ)
p(ω) = −
(1 + λ − λCIo
2 (ω))
q(ω) = − c(1 + λ)
(1 + λ − λCIo2 (ω)) (14)
G2 (ω|p, q) − CIo
2
(ω)V 2 (ω|p, q) = 0
λ > 0.
By solving the third of (14), one gets the roots
1 + CIo (ω) 1 − CIo (ω)
λ1 = 2 (ω) − 1 , λ2 = 2 (ω) − 1 . (15)
CIo CIo
2
Since CIo (ω) > 1 and the Lagrangian multiplier must be positive, the second root
is definitely discarded and the first two of (14), with λ = λ1 , yield the control
parameters that comply with the optimal criterion expressed by (9)
k − mω 2 c
p(ω) = ; q(ω) = . (16)
CIo (ω) − 1 CIo (ω) − 1
Assuming that c > 0, the coefficient
p q(ω) is always positive, while p(ω) is negative
for values of ω larger than ωo2 = k/m.
Note that for ω = ωo , p(ω) = 0 and at this pulsation the control is exclusively
demanded to the derivative term. This result agrees with the fact that the deriva-
tive linear term turns out to be prominent in any control algorithm, especially at
resonance [Baratta and Di Paola, 1996]. Note anyway that this is not a choice, but
it is the optimal solution of the conditioned optimization problem in Eq. (9).
Note that if Co (ω) is introduced as a even function of ω, also λ1 is a even
function, and both the control coefficients p(ω) and q(ω) are real and even functions
of ω.
one gets
Z +∞ Z +∞
1 1
m√ −ω uo (ω)e
2 jωt
dω + c √ jωuo (ω)ejωt dω
2π −∞ 2π −∞
Z +∞ Z +∞
1 1
+k√ uo (ω)ejωt dω + √ p(ω)uo (ω)ejωt dω
2π −∞ 2π −∞
Z +∞ Z +∞
1 1
+√ jωq(ω)uo (ω)e dω = √
jωt
fo (ω)ejωt dω (18)
2π −∞ 2π −∞
whence one concludes that
Z +∞ Z +∞
1 1
u(t) = √ uo (ω)e dω = √
jωt
u(ω, t)dω (19)
2π −∞ 2π −∞
with
Z +∞ Z +∞
1 1
u̇(t) = √ u̇(ω, t)dω = j √ ωuo (ω)ejωt dω
2π −∞ 2π −∞
Z Z (20)
+∞ +∞
1 1
ü(t) = √ ü(ω, t)dω = − √ 2
ω uo (ω)e jωt
dω
2π −∞ 2π −∞
yielding the response to the excitation f (t) of the structure, controlled by the active
force
Z +∞ Z +∞
1 1
w(t) = √ p(ω)uo (ω)ejωt dω + √ jωq(ω)uo (ω)ejωt dω
2π −∞ 2π −∞
Z +∞ Z +∞
1 1
= √ p(ω)u(ω, t)dω + √ q(ω)u̇(ω, t)dω = wu (t) + wu̇ (t) .
2π −∞ 2π −∞
(22)
For a generic bounded-support forcing function (like in case of seismic action) the
existence of all the above transforms is ensured, and uo (ω) and u(t) constitute a
Fourier transform pair
Z +∞ Z +∞
1 1
uo (ω) = √ u(τ )e−jωτ dτ ; jωuo (ω) = √ u̇(τ )e−jωτ dτ . (23)
2π −∞ 2π −∞
It is possible to manipulate the algorithm yielding the control force in Eq. (22).
In fact, considering the inverse Fourier transforms of p(ω) and q(ω)
Z +∞ Z +∞
1 1
P (x) = √ p(ω)ejωx dω ; Q(x) = √ q(ω)ejωx dω (24)
2π −∞ 2π −∞
July 23, 2002 10:28 WSPC/124-JEE 00071
introducing Eqs. (23) into (22) and distinguishing the two terms wu (t) and wu̇ one
can write
Z +∞ Z +∞ Z +∞
1 1 −jωτ
wu (t) = √ jωt
p(ω)uo (ω)e dω = p(ω) u(τ )e dτ ejωt dω
2π −∞ 2π −∞ −∞
Z +∞ Z +∞
1 1
= √ √ p(ω)e jω(t−τ )
dω u(τ )dτ
2π −∞ 2π −∞
Z +∞
1
= √ P (t − τ )u(τ )dτ (25)
2π −∞
Z +∞
1
wu̇ (t) = √ jωq(ω)uo (ω)ejωt dω
2π −∞
Z +∞ Z +∞
1 −jωτ
= q(ω) u̇(τ )e dτ ejωt dω
2π −∞ −∞
Z +∞ Z +∞
1 1
= √ √ q(ω)e jω(t−τ )
dω u̇(τ )dτ
2π −∞ 2π −∞
Z +∞
1
= √ Q(t − τ )u̇(τ )dτ . (26)
2π −∞
Note that P (x) and Q(x) are real and even functions of x, in that both p(ω) and
q(ω) are real and even functions of ω.
One should remark that the response u(t) [Eq. (19)] results from the superposi-
tion of harmonic components, each optimised in the sense of the problem (9), with
the relevant component of the control obeying the constraint inequality. Hence the
final control force remains bounded by
Z Z +∞
1 +∞ 1
|w(t)| = √ w(ω|p, q)e −jωt
dω ≤ √ Co (ω)|fo (ω)|dω . (27)
2π −∞ 2π −∞
Whence one understands how the function Co (ω) can be properly designed in such
a way to have the desired threshold level of the control force.
With reference to the practical implementation of the proposed control
algorithm, note that any control rule requires some prediction; this can, then, be
reviewed as a common problem, met even in the optimal calibration of control coef-
ficients in classical linear control that is nevertheless widely applied (see i.e [Soong,
1990; Chap. 3. Baratta and Corbi, 2000]). So, practical feedback implementation
will be forced to accept some manipulation and/or approximation. The problem
is under current development and it is not yet fully afforded in this paper, which
is intended to set up the basis of the method and to test its effectiveness in the
most favourable situation, when the Fourier decomposition of the forcing function
is known.
July 23, 2002 10:28 WSPC/124-JEE 00071
r Z +∞
2
u(t) = [uoR (ω) cos ωt − uoI (ω) sin ωt]dω (30)
π 0
Z +∞
1
u̇(t) = j √ ωuo (ω)ejωt dω
2π −∞
Z +∞
1
= √ ω[juoR (ω) − uoI (ω)] [cos ωt + j sin ωt]dω
2π −∞
r Z +∞
2
=− ω[uoR (ω) sin ωt + uoI (ω) cos ωt]dω (31)
π 0
Z +∞
1
ü(t) = − √ ω 2 uo (ω)ejωt dω
2π −∞
Z +∞
1
= −√ ω 2 [uoR (ω) + juoI (ω)] [cos ωt + j sin ωt]dω
2π −∞
r Z +∞
2
=− ω 2 [uoR (ω) cos ωt − uoI (ω) sin ωt]dω (32)
π 0
July 23, 2002 10:28 WSPC/124-JEE 00071
Z +∞
1
wu (t) = √ p(ω)uo (ω)ejωt dω
2π −∞
r Z +∞
2
= p(ω)[uoR (ω) cos ωt − uoI (ω) sin ωt]dω
π −∞
Z +∞
1
wu̇ (t) = √ jωq(ω)uo (ω)ejωt dω
2π −∞
r Z +∞
2
=− ωq(ω)[uoR (ω) sin ωt + uoI (ω) cos ωt]dω (33)
π 0
r Z +∞ r Z +∞
2 2
P (x) = p(ω) cos ωxdω ; Q(x) = q(ω) cos ωxdω (34)
π 0 π 0
r Z +∞ r Z +∞
2 2
wu (t) = P (t − τ )u(τ )dτ ; wu̇ (t) = Q(t − τ )u(τ )dτ (35)
π 0 π 0
Since the initial conditions at t = 0 for u(ω, t) are
[k + p(ω) − mω 2 ] ω[c + q(ω)]
u(ω, 0) = uo (ω) = −j 2 fo (ω)
G2 (ω|p, q) G (ω|p, q)
2 (36)
ω [c + q(ω)] [k + p(ω) − mω 2 ]
u̇(ω, 0) = jωuo (ω) = + jω fo (ω)
G2 (ω|p, q) G2 (ω|p, q)
the particular solution u(t) obeys the following initial conditions
r Z +∞ r Z +∞
2 2 [k + p(ω) − mω 2 ]
u(0) = uoR (ω)dω = fo (ω)dω
π 0 π 0 G2 (ω|p, q)
r Z +∞ r Z +∞ (37)
2 2 [c + q(ω)]
u̇(0) = − ωuoI (ω)dω = ω2 2 fo (ω)dω
π 0 π 0 G (ω|p, q)
a)
150
Minimum constrained
point H decreasing H=const.
100 Co=0.5
Co<0.5
50
0
0 200 400 600 800 1000
p
b) -100
H decreasing
-200
400 q
Minimum constrained
point
200
Co <0.4
0 p
-2000 -1000 0 1000 2000 3000
c) Co =0.4
-200 H decreasing
-400
Fig. 1. (a) k − mω2 = 0: the optimal point is always on the q-axis; (b) k − mω2 < 0: at the
optimal point the product p · q is always negative; (c) k − mω2 > 0: at the optimal point p and q
are both positive.
July 23, 2002 10:28 WSPC/124-JEE 00071
can be unstable or, at least, trigger somewhat different responses with varying the
initial conditions.
In order to illustrate this effect, consider the structure acted on by the harmonic
force
fψ (t) = a1 sin ψt + a2 cos ψt . (38)
The optimally controlled (in the sense of Sec. 2.1) structure response obeys the
equation
mü(ψ, t) + c0 (ψ)u̇(ψ, t) + k 0 (ψ)u(ψ, t) = fψ (t) (39)
where
k 0 (ψ) = k + p(ψ) ; c0 (ψ) = c + q(ψ) . (40)
The particular solution in the form equivalent to (4) is
u(ψ, t) = u1 (ψ) sin ψt + u2 (ψ) cos ψt
(41)
u̇(ψ, t) = ψ[u1 (ψ) cos ψt − u2 (ψ) sin ψt]
with
[k 0 (ψ) − mψ 2 ]a1 + ψc0 (ψ)a2
u1 (ψ) = ;
[k 0 (ψ) − mψ 2 ]2 + c0 (ψ)2 ψ 2
(42)
[k 0 (ψ) − mψ 2 ]a2 − ψc0 (ψ)a1
u2 (ψ) = .
[k 0 (ψ) − mψ 2 ]2 + c0 (ψ)2 ψ 2
Equation (41) will be referred to as the stationary solution; it obeys the following
initial conditions at t = 0
u(ψ, 0) = u2 (ψ)
(43)
u̇(ψ, 0) = ψu1 (ψ) .
Equation (42) proves that the stationary solution remains stable provided that
both k and c in the original equation are larger than zero, as can be easily verified
considering that p(0) > 0, q(ψ) > 0 ∀ψ [see (16)].
By contrast, let consider the general integral of Eq. (39). The characteristic
equation is
mθ2 + c0 (ψ)θ + k 0 (ψ) = 0 (44)
whose roots are
p p
−c0 (ψ) + c0 (ψ)2 − 4k 0 (ψ)m −c0 (ψ) − c0 (ψ)2 − 4k 0 (ψ)m
θ1 = ; θ2 = (45)
2m 2m
and the general integral of Eq. (39) is
U (t|A, B) = Aeθ1 t + Beθ2 t + u(ψ, t)
(46)
U̇ (t|A, B) = Aθ1 eθ1 t + Bθ2 eθ2 t + u̇(ψ, t) .
July 23, 2002 10:28 WSPC/124-JEE 00071
1E-2
2 E+ 0
..
u(t)
w(t),ug(t)
t
0 E+ 0
8E-3
0.00
0.40
0.80
1.20
1.60
-2E+0
4E-3
base acceleration
-4E+0
0E+0
-6E+0
0.00 0.40
-4E-3 -8E+0
(a) (b)
Fig. 2. k 0 (ψ) < 0: structural time response (a), control action and forcing function (b).
July 23, 2002 10:28 WSPC/124-JEE 00071
1E-2
%
2E+0
%
..
u(t) w(t),ug(t)
0E+0
8E-3 0.00 0.40 0.80 1.20 1.60
& '
(
)
*
'
*
*
'
*
*
'
4E-3 + *
base acceleration
& '
-4E+0
0E+0 t
%
-6E+0
0.00 0.40 0.80 1.20 1.60
-4E-3 -8E+0
(a) (b)
Fig. 3. k 0 (ψ) = 0: structural time response (a), control action and forcing function (b).
2E+0
1
<
.. <
2E-3
%
u(t) ( * * * * *
w(t),ug(t) ,
2
4 4
3
2 4
5
9 9 9 9 9
3 3 3 3 3
2 6 4 8 8 2 4 8 4 8
base acceleration
3
; 2 2 ; ; 4
1E-3 1E+0
t
1 1
0E+0
%
t 0E+0
1 1 1 1 1 1 1 1 1
0.00 0.40 0.80 1.20 1.60 0.0 0 0.40 0.80 1.20 1.60
-1E+0
/
-1E-3
-2E+0
. 1
/
-2E-3
(a) (b)
Fig. 4. k 0 (ψ) > 0: structural time response (a), control action and forcing function (b).
(p(ψ) = −k and q(ψ) = 0.5 kg·s/cm are now considered) the homogeneous response
is bounded but it takes values much higher than the stationary one [Fig. 3], while
Fig. 4 (where one assumes p(ψ) = −675 kg/cm and q(ψ) = 0.375 kg·s/cm) proves
that, even if the system remains stable (k 0 (ψ) > 0), the homogeneous response can
be very different from the stationary one, for a rather long transient time (t ≈ 8To ).
If one looks at the control force, one realizes that in the transient phase a surplus
control may be required, disturbing the optimal character of the algorithm.
u(t)
2E+0
..
w(t),ug(t)
8E-4
base acceleration
1E+0
4E-4
0E+0
t 0E+0 t
0.00 0.40 0.80 1.20 1.60 0.00 0.40 0.80 1.20 1.60
-1E+0
-4E-4
-8E-4 -2E+0
(a) (b)
Fig. 5. Response (a) and control force (b) by including “norm control” with optimal coefficients
k̄ = 572.2 kg/cm, c̄ = 32.3 kg·s/cm, obtained for Wo = 1/4 of the forcing function energy.
July 23, 2002 10:28 WSPC/124-JEE 00071
2E-3
u(t)
%
2E+0
Ï
..
norm controlled
&
w(t),ug(t)Á
É Ç
Ê
uncontrolled
Ê Ê È
u nco ntrolled
Æ
Ë Å Ç Ë É Ç Î
Ú
b ase acceleration
Æ
Î Å Å Î Î Ç
1E-3
1E+0
0E+0 t
0E+0
Ã
Ä
t
0.00 0.40 0.80 1.20 1.60
Û Â
-1E-3
-1E+0
Ã
-2E-3 -2E+0
(a) (b)
Fig. 6. Response (a) and control force (b) only by “norm control”, with optimal coefficients
k̄ = 572.2 kg/cm, c̄ = 32.3 kg·s/cm, obtained for Wo = 1/4 of the forcing function energy.
Sa (cm 2 sec -2 )
2,50
f(t)
. sec −1
300
Ω a = 387
2,00
200 ξ a = 41%
1,50
100 Ω b = 18.37 sec −1
1,00
t
ξ b = 37%
0
0,50
-100
-200 0,00
0,09 8,97 17,85 26,74 35,62 44,50 53,38 62,27
-300
ω(sec -1 )
0 4,88 9,76 14,6 19,5 24,4 29,3 34,2 39 43,9 48,8 53,7 58,6 63,4 68,3
(a) (b)
Fig. 7. The Sturno earthquake time-history (a) and power spectrum (b).
1.00
1.20
p (ω)/max|p (ω)| |Ho(ω)|/max|Ho(ω)|
Co(ω) |H (ω)|/|Ho(ω)|
0.80
w(ω|p,q)/fo(ω)
0.80
0.60
0.40 0.40
0.20
ω
0.00
0.00 20.00 40.00 60.00 80.00 100.00
-0.40
(a) (b)
1E+0
u(t) 2E+2
..
w(t),ug(t)
" "
unctrl.d displacement
! !
# $ $
base acceleration
" "
ctrl.d displacement
! !
# $ $
ctrl force
5E-1
1E+2
0E+0 0E+0
-5E-1 -1E+2
-1E+0
t -2E+2 t
(a) (b)
Fig. 9. Structural time response (a), control action and forcing function (b) for the Sturno
earthquake.
July 23, 2002 10:28 WSPC/124-JEE 00071
where ρ is a normalisation factor, yielding the equality of the total energy between
the actual and the smooth spectrum [Baratta and Zuccaro, 1999], and the param-
eters Ωa , ξa , Ωb , ξb are given in Fig. 7(b).
The control strategy can be designed in way that the control action is more
intense at the frequency that can transmit more energy to the structure. This
scope can be achieved giving the function Co (ω) the following expression
(
Com S(ω)|Ho (ω)| if Com S(ω)|Ho (ω)| ≤ Co max
Co (ω) = (56)
Co max otherwise .
In Figs. 8 and 9, the results of the simulation are quoted assuming Com =
Co max = 0.9.
5. Conclusions
In the paper a procedure, able to optimise the control action with respect to the
foreseen forcing function and constrained by an upper bound on the control action,
has been developed. The procedure is based on the harmonic decomposition of
the forcing function and on the ideal distribution of the control action over the
frequency range in an optimal fashion with respect to the expected power density
distribution of the active force. The set up of the rationale allows to distribute the
control power over the frequency range, by simply assigning the function Co (ω),
in way to save the control power where the control action is expected to be less
effective. Furthermore the significant influence of the initial conditions on what one
could deduce if referring to the frequency domain has been deeply analysed. A wide
numerical investigation (here only partially reported) has been finally carried out,
demonstrating the effectiveness of the procedure, that allows a consistent response
mitigation with a minimum control energy employ.
References
Baratta, A. and Di Paola, M. [1996] “Optimal Non-Linear Control Law for Linear
Sdof Structures,” Proc. 1st European Conference on Struct. Control, Barcelona,
pp. 56–63.
Baratta, A., Cennamo, C. and Voiello, G. [1998] “Design of linear active control algorithms
for non-stationary unknown forcing function,” J. Struct. Control 2, pp. 7–31.
Baratta, A. and Corbi, O. [1999] “Improved control of structures by time-delayed algo-
rithms,” Proc. of the 5th International Conference on the Application of Artificial
Intelligence to Civil and Structural Engineering, Oxford, pp. 159–168.
Baratta, A. and Corbi, O. [1999] “Algoritmo di controllo per forzanti armoniche nel
dominio delle frequenze,” Proc. of the 12nd Italian Congress of Computational
Mechanics, GIMC, Napoli, pp. 11–16.
Baratta, A. and Corbi, O. [1999] “Controllo attivo di sistemi strutturali mediante algo-
ritmo frequenziale combinato,” Proc. of the 14th Italian Congress of Theorical and
Applied Mechanics, AIMETA, Como.
July 23, 2002 10:28 WSPC/124-JEE 00071
Baratta, A. and Corbi, O. [2000] “On the optimality criterion in structural control,”
Earthq. Engg. Struct. Dyn. 29, pp. 141–157.
Baratta, A. and Zuccaro, G. [1999] “How to utilize the anti-optimization analysis to treat
uncertainty in seismic excitation?,” in “Whys and Hows in Uncertainty Modeling.
Probability, Fuzziness and Anti-optimization” I. Elishakoff Ed., CISM Courses and
Lectures, 388, Springer-Verlag, pp. 1–59.
Housner, G. W. and Masri, S. F. [1996] “Structural control research issues arising out
of the Northridge and Kobe earthquakes,” Proc. of the 11th World Conference on
Structural Control, Acapulco, paper n.2009.
Housner, G. W., Bergman, L. A., Caughey, T. K., Chassiakos, A. G., Claus, R. O., Masri,
S. F., Skelton, R. E., Soong, T. T., Spencer Jr., B. F. and Yao, T. P. [1997] “Structural
control: Past, present, and future,” Eng. Mech. ASCE 123(9), pp. 897–971.
Kobori, T. [1998] “Mission and perspective towards future structural control research,”
Proc. of the 2nd World Conference on Structural Control, Kyoto, 1, pp. 25–34.
Soong, T. T. [1990] “Active Structural Control: Theory and Practice,” Longman, England
and Wiley, New York, pp. 1–194
Soong, T. T., Reinhorn, A. M., Aizawa, S. and Higashino, M. [1994] “Recent structural
applications of active control technology,” J. Struct. Control 1(2), pp. 5–21.