You are on page 1of 9

OPTIMAL CONTROL OF ADAPTIVE/SMART BRIDGE STRUCTURES

By H. Adeli! and A. Salehz

ABSTRACT: Through the use of active controllers, a structure can modify its behavior during dynamic loadings
such as impact, wind, or earthquake loadings. Such structures with self-modification capabilities are called smart
structures. The smart-structure technology will have enormous consequences in terms of preventing loss of life
and damage to structure and its content specially for large structures with hundreds of members. In this paper,
a computational model is presented for active control of large adaptive structures subjected to dynamic loadings
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

such as impact, wind, and earthquake loadings. The governing differential equations of the open-loop and closed-
loop systems are formulated, and a recursive approach is presented for computing the response of the structure.
A robust parallel-vector algorithm is developed for the recursive solution of the response of the open-loop and
closed-loop systems. The computational model is applied to active control of large bridge structures. Three
different schemes are investigated for optimal placement of controllers in bridge structures. Results are presented
for three types of bridge structures: single-span, multispan continuous, and curved steel-truss bridges.

INTRODUCTION consider linear actuators only. Linear actuators are usually po-
sitioned at member ends (Fig. 2). One actuator placed at one
Active control of structures has been recognized as one of end of the member is usually sufficient. However, when the
the most challenging and significant areas of research in struc- force exertion capacity of one actuator is not enough, two ac-
tural engineering in recent years (Housner et al. 1996; Kobori tuators may be used, one at each end of the member.
1996). Through the use of active controllers, a structure can For large structures with hundreds of members, the sensor/
modify its behavior during dynamic loadings such as impact, actuator placement for optimum performance is an important
wind, or earthquake loadings. Such structures with self-mod- and challenging problem. The approaches that have been used
ification capabilities are called smart structures. The smart- in the literature are based on selecting an arbitrary number of
structure technology can revolutionize the structural engineer- actuators, placing them at different locations in the structure
ing profession. It will have enormous consequences in terms and measuring a performance criterion, usually in terms of the
of preventing loss of life and damage to structure and its con- total energy stored in the control system (Arabel 1981; Schulz
tent specially for large structures with hundreds of members. and Heimbold 1983, Sakamoto et a1. 1992; Furuya and Haftka
For example, through active control of structures one can limit 1995). With the exception of Furuya and Haftka (1995), all
the deformation of a high-rise building structure to the linear the published articles deal with small structures. Furuya and
range, thus preventing any permanent structural and nonstruc- Haftka (1995) report that the number of possible combinations
tural damage, and eliminating the damage to its content that even for small structures can be very large. For example, using
may be substantially more valuable than the structure itself. eight actuators they found the number of candidate sets of
The research published in this area so far is limited to small- locations for a truss of 1,507 members to be 648 X lOt8.
scale structures with a few members, such as the experimental In this article, we present a computational model for active
work by Subramaniam et al. (1996) and Dyke et al. (1996), control of large structures subjected to dynamic loadings such
and analytical work by Oz et al. (1990), Yang et al. (1992), as impact or earthquake loadings. We use the same numbers
and Khot (1994). Our research goal is to develop computa- of sensors and actuators collocated at member ends and dis-
tional models for active control of large adaptive/smart struc- tributed throughout the structure. The governing differential
tures with hundreds of members.
In a smart structure, we design a predetermined number of
members to be actively controlled members. Each such mem-
ber has a sensor and an actuator. The sensor measures the
displacement along the degrees of freedom, and the actuator
applies the required force for the appropriate correction to the
uncontrolled response that is determined by signal conditioners
(devices that translate the output signals from the sensors to
forces to be exerted by the actuators) in the state-feedback
control system (Fig. 1).
Actuators are devices that can produce given forces or
strains. Actuators that produce forces in the axial direction are
called linear actuators. The driving-force mechanism in linear
actuators can be hydraulic, electric, or electrohydraulic. Ac- ~-------! G '- -.J

tuator forces can range from a few Newtons (pounds) to a few


hundred kilo Newtons (kilo pounds). In the present work, we
'Prof., Dept. of Civ. Engrg., The Ohio State Univ., 470 Hitchcock Hall,
2070 Neil Ave., COlumbus, OH 43210-1275.
'PhD Candidate, Dept. of Civ. Engrg., The Ohio State Univ., 470
0 summation

Hitchcock Hall, 2070 Neil Ave., Columbus, OH.


Note. Associate Editor: Kevin Z. Truman. Discussion open until July (> integration
I, 1997. To extend the closing date one month, a written request must
be filed with the ASCE Manager of Journals. The manuscript for this
paper was submitted for review and possible publication on March 13,
1996. This paper is part of the Jour",,' of Structural Engineering, Vol.
0 computation

123, No.2, February, 1997. ©ASCE, ISSN 0733-9399/97/0002-0218- FIG. 1. State Feedback Control System (Variables Are Defined
0226/$4.00 + $.50 per page. Paper No. 12886. In AppendiX III)

218/ JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997

J. Struct. Eng., 1997, 123(2): 218-226


A matrix obtained by minimizing the control performance index

"':~~
J

(4)
Supporting ban
given by
Transverse beam G = R-lBTp (5)
where P =2N X 2N positive definite matrix called the Riccati
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

matrix obtained from the solution of the following Riccati


equation (Saleh and Adeli 1994b):
Q + PA + ATp - PBR-IBTp = 0 (6)

In this equation, Q = 2N X 2N state weighting matrix; and


R = 2N X 2N control weighting matrix. Weighting matrix Q
specifies the relative importance of the various components of
FIG. 2. Example of P081tlonlng Actuator In Tru88-Brldge !he state vector X. For example, if XI is of concern while X2
Structure
IS not, we may choose a state weighting matrix

equations of the open-loop and closed-loop systems are for-


mulated, and a recursive approach to compute the response of
Q = [~ g] (7)

the structure is presented. The solution of the governing dif- If both Xl and X2 are of equal concern, we may choose a state
ferential equations for large structures requires an enormous weighting matrix
amount of computer processing time and can be achieved only
= [~ ~]
on high-performance computers (Adeli 1992a,b) within a rea-
sonable amount of time. Thus, we present a robust parallel- Q (8)
vector algorithm for the recursive solution of the response of
the open-loop and closed-loop systems. For large structures In this research, we choose the state weighting matrix Q as ql
the major bottleneck in this problem is the solution of the where I is a 2N X 2N identity matrix and q is a weight factor.
complex eigenvalue problem encountered in the solution of Similarly, the control weighting matrix R is chosen as rl,
the resulting Riccati equation as well as the solution of both where r is a weight factor. The effects of weight factors q and
open-loop and closed-loop systems of equations. The methods r on the co~trolled response of the structure are investigated.
reported in the literature yield satisfactory results for small The solutIon for the state variable vector, X, in the closed-
p~oble~s but often become unstable for large problems (Gar-
loop system [(2)] can be expressed as (Casti 1987)
d1Oer, 10 press, 1997). Recently, the writers have developed X(l+1) =nx(k) + rr~), k = 1, 2, . . . (9)
robust parallel algorithms for solution of the complex eigen-
value problem of an unsymmetric matrix and applied them where the 2N X 2N matrix n is given by
successfully to large matrices (Saleh and Adeli 1996). n = exp(8TA) =e0e'T (10)
We investigate three different schemes for placement of
controllers in bridge structures and apply the computational The 2N X Nt. matrix r is given by
model to active control of three large bridge structures.

RESPONSE OF CONTROLLED STRUCTURES


r= [fT eXP(8TA)dt] B = eae'TB o o (11)

The discretized differential equation governing the uncon- and the 2N X 2N matrices 0 and a are given by
~olled. m~tion of a structure, the open-loop system of equa-
tIOns, IS gIven by (Saleh and Adeli 1994b) o = exp(8TA) (12)
and
X =AX + BoCo (1)
&T

where X = 2N state variable vector; A = 2N X 2N plant matrix;


Bo = 2N X Nt. applied load-distribution matrix, which relates
the .applied external dynamic forces to the structure global co-
a=
L
0 exp(8TA)dt (13)

ordmates system; and (, = vector of applied dynamic loading. where A = diagonal matrix whose nonzero diagonal elements
The .number of e.lements in Co, Nt. is equal to number of dy- are the eigenvalues of matrix A; 8T is the time increment· e
namIC loads applIed at predetermined displacement degrees of = matrix of right eigenvectors obtained from the solution 'of
freedom. = =
Ae Ae; and e' matrix of left eigenvectors obtained from
The equation governing the controlled motion of a structure, the solution of e'A = Ae'.
the closed-loop system of equations, is given by (Saleh and Matrix 0 can be expressed as
Adeli 1994b) 2
0.(8T) =I + 8TA + 8T A2 + ... + 8T' A' (14)
X =Ax + BoCo (2) 2! n!

where A = 2N X 2N unsymmetric closed-loop matrix given ~here. n = n?~ber of te~s u~ed to compute the series expan-
~Ion (10 addI~Ion to the IdentIty matrix term). The expression
by
10 (~4) !equIres n(n - 1)/2 diagonal-matrix multiplications
A =A - BG (3) (M~Irov~tch 1985). To reduce the number of matrix multipli-
cation~, 1Oste~ of (14) we use a recursive equation to compute
where B = 2N X Nt input matrix; and G = optimum-gain 0, WhICh requIres only (n - 1) matrix multiplications
JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997/219

J. Struct. Eng., 1997, 123(2): 218-226


L
SchemeA I

Scheme B

SchemeC
Cross beam
II
U2

U3
}- planes

Bridge deck
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

18 x 2.7 m ~48.6 m.

FIG. 3. Example 1-0ne-Span Truss Bridge

Scheme B I 1
L
Scheme A ,---,----r-·r.·,·-.-I---'1-,.-,.....,--
U2
I
U3
I
1

i
I

I
r--r--r I
L

U2

U3
I

I
I

I
1

I
~

I
j Controller
planes

Scheme C I I 1 I I I

~I
12 x 3 m. = 36 m. t2 x 3 m. = 36 m.

FIG. 4. Example 2-lWo-Span Trus. Bridge

= 1 + 8TA"_1 (15) SCHEMES FOR PLACEMENT OF CONTROLLERS


'I =
8i8T)
where 2N X 2N matrix defined by Where should the controllers (actuators and sensors) be
8TA placed for optimum control of response of a given structure?
'1=1 +n- (16) This is a significant question specially for large structures with
hundreds of members. Only a few researchers have attempted
and to answer this question. We already mentioned the work of
Furuya and Haftka (1995) in the introduction. Stubbs and Park
y _ 1 8TA (17) (1996) propose simple rules for the placement of sensors (ac-
~I - +n+ 1- i '1-10 i = 2,3, ... , n
celerometers) to regenerate mode shapes of simply supported
The series expansion of (17) will yield (14). Thus, to obtain single-span and continuous two-span beams using Shannon's
the response of the closed-loop structure from (9) we first need sampling theorem (Marks 1993). Their presentation, however,
to find the comylex right and left eigenvectors of the unsym- does not extend to the placement of actuators to control the
metric matrix A, which is covered in Saleh and Adeli (1996). response of the structure.
Then, the response is obtained through the application of the Large bridge structures nonnally have a significant stiffness
recursive (15)-(17). in their longitudinal direction and consequently have substan-
The solution for the state variable vector, X, in the open- tial inherent resistance to carry dynamic loadings, say, from
loop system, [(1)] is obtained similarly by replacing matrix A earthquakes or winds, in the longitudinal direction without
by matrix A in (10) and (11). failure (similar to the inherent stiffness of multistory building
The external dynamic-force vector, f o , can be in different structures in the vertical direction). In the transverse as well
fonns such as earthquake loading, wind loading, or impulse as the vertical directions, however, bridges are quite vulnerable
loading. In the case of earthquake loading, f o is represented to dynamic loadings. Consequently, we investigate schemes
by - MUg, where M = structure mass matrix; ug = ugl; and ug for placement of controllers in order to minimize the response
= ground acceleration recorded on an accelerograph (acceler- of the bridge in the transverse and vertical directions or in the
ogram). In the case of impulse loading, f o is in the fonn P8(t), plane of the bridge cross section. In the unlikely event when
where P = vector of magnitudes of impulsive forces applied vibrations in the longitudinal directions need to be controlled
at predetennined displacement degrees of freedom of the struc- also, the required control forces will be minimal and can be
ture; and 8(t) = delta function. Wind loading on a structure handled by installing just two controllers, one at each support.
can be represented as a series of impulse loadings. The dynamic response of a bridge structure in the transverse
Using the vectorization and microtasking capability of the and vertical directions is mostly due to the first few modes of
shared-memory supercomputers such as the Cray YMP-8/8128 vibration (usually the first two or three modes of vibrations)
(Saleh and Adeli 1994a), we have developed an efficient par- (Chopra 1995). To control the vibrations of the bridge in the
allel-vector algorithm for the recursive computation of the re- plane of the bridge cross section we propose to place four
sponse of the open-loop and closed-loop systems, which is controllers in the plane of the cross section and investigate
presented in Appendix I. three different schemes. The findings of this research are ap-
220 I JOURNAL OF STRUCTURAL ENGINEERING I FEBRUARY 1997

J. Struct. Eng., 1997, 123(2): 218-226


L
I

~~,o
SchemeA

U2
SchemeB
planes
U3
SchemeC
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

FIG. 5. Example 3-0n8-Span Curved Truss Bridge

plicable to any kind of bridges consisting of two identical pla- designed for the American Association of State Highway and
nar structures connected to each other at the top and bottom. Transportation Officials (AASHTO) live load of H20
However, we limit our examples to three kinds of steel truss (AASHTO 1993) and according to the American Institute of
bridges: single-span (Fig. 3), multiple-span continuous (Fig. Steel Construction (AISC) Allowable Stress Design (ASD)
4), and curved bridges (Fig. 5). Each bridge consists of two specifications (AISC 1989). Wide-flange shapes are selected
steel trusses, one on each side of the bridge. In a plane of for all the members of the bridge structure using A36 steel
cross section two controllers are placed vertically along the with yield stress of 248.2 MPa (36 ksi).
vertical member of each truss, and two controllers are placed Then, three kinds of dynamic loadings are considered
horizontally along the horizontal members connecting the two
trusses at the top and bottom. There are also cross bracings in
both lower and upper horizontal planes of the structure in the 1. A moving vertical impulse loading on each lane of the
case of straight bridges (Figs. 3 and 5) and on the top of the bridge moving in the same direction with a speed of 65
structure in the case of the curved bridge (Fig. 4). mph (105 km/h). The magnitude of this load is the re-
In all the three schemes the same numbers of actuators and sultant of the two axle loads of the AASHTO H20 load-
sensors are collocated along the actively controlled members ing. This load is multiplied by a 8 function with a du-
in the vertical cross-sectional planes of the bridge. In scheme ration of 1 s.
A four controllers are placed in every vertical plane of the 2. The 1940 El Centro earthquake ground-acceleration rec-
truss passing through the joints except the end planes where ord (Fig. 6).
only one controller is placed horizontally along the member 3. Periodic impulsive horizontal loadings on each joint of
connecting the two trusses at the top. the truss (perpendicular to the plane of the truss) mod-
In scheme B controllers are placed in the vertical planes eling the wind loading on the structure (pressure on one
passing through the joints over the middle half of each span side and suction on the other side), as shown in Fig. 7.
of the bridge when the number of panels in each span is a The same magnitude is used for all the loads. The mag-
multiple of four. Otherwise, the middle half of each span is nitude of the wind pressure is found using the AASHTO
extended from each end up to the adjacent vertical planes of (1993) code and assuming a wind velocity of 160 km/h
members passing through the joints (for example, the bridge (100 milh): q = 3.59 kN/m2 • The bridge structure is re-
in Fig. 3 has 18 2.7 m panels, and we include controllers in designed for one of the following load combinations sat-
the vertical planes of 10 panels). This arrangement is inves- isfying all the AISC, ASD, and AASHTO stress re-
tigated because it covers the locations of the maximum re- quirements: a + b or a + c. For wind loading the
sponses of the first two modes of vibrations. AASHTO displacement constraint is also satisfied. The
In scheme C controllers are placed in the vertical planes maximum vertical displacement due to the traffic loading
passing through the joints over the middle third of each span and the maximum horizontal displacement due to the
of the bridge when the number of panels in each span is a wind loading are limited to Lll,OOO, where L is the span
multiple of six. Otherwise, the middle third of each span is length. The resulting structure is the open-loop system.
extended from each end up to the adjacent vertical planes of
members. This arrangement does not cover the locations of Next, controllers are added using one of the three schemes
the maximum response of the second mode of vibrations. In- A, B, and C. The response of the closed-loop system is then
vestigation of this arrangement will shed light on the signifi- compared with the response of the open-loop system for the
cance of the participation of the second mode in the controlled aforementioned dynamic-loading cases. We also considered
structure. combination of traffic loading with wind and earthquake load-
ing for the controlled structure (a + b or a + c). Since the
EXAMPLES traffic loading is in the vertical direction and wind and earth-
Three different example bridge structures are investigated. quake loadings are primarily in the horizontal direction, we
In all the examples, it is assumed that the bridge deck consists found negligible interactions between the responses of the con-
of a 7 in. (17.78 cm) concrete deck. Each bridge is initially trolled structure in the horizontal and vertical directions. Con-
JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997/221

J. Struct. Eng., 1997, 123(2): 218-226


0.4
structure has 388 members and 100 nodes. This results in 564
state variables (three displacements and three velocities for
0.3 each node of the structure). The three schemes A, B, and C
for placement of the controllers are identified in Fig. 4. The
0.2 control weighting factor r is chosen to be equal to 0.1.
B Example 3
.g 0.1
This is a single-span curved truss bridge with a span of 48.6

i 0 m, a height of 6 m, and a width of 6 m (Fig. 5) (the same as


example 1). 1\\'0 different values are chosen for height H in
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

1 Fig. 5: 2 and 6 m. The structure has 292 members and 76


~
-0.1
nodes. This results in 432 state variables (three displacements
and three velocities for each node of the structure). The three
-0.2 schemes A, B, and C for placement of the controllers are iden-
tified in Fig. 5. The control weighting factor r is chosen equal
-0.3 to 0.1.
NUMERICAL RESULTS
-0.4
0 5 10 15 20 2S 30 In this section numerical results are presented for three ex-
ample bridges described in the previous setion using three dif-
TIme (Sec.) ferent schemes for actuator placements A, B, and C.
FIG. 6. 1940 EI Centro (California) Earthquake Ground Accel-
erations e 2.5 r
~ 2.0
'l!
-5 1.5
~
f! 1.0
~
.5 0.5 -Open loop
8-
1 0.0
-r-t.O
-kr--t--tl~+'~<:±"=lf>-'9"'\--<+--+1r----+---'l-~~ - - - r-0.5
.s
';i -0.5
·--···r-O.l

5
! -l.0
i
:; -1.5

~ -2.0
~
-2.5

FIG. 8. Vertical Displacement at Midpoint of Top Chord of


o 5 10 15 20 2S 30 Tru.. of Example 1 under Traffic Loading, Using Scheme A for
Both Open-Loop System (Uncontrolled Structure) and Closed-
TIme (Sec.) Loop System (Controlled Structure) Using Three Different Val-
ues for ContrOl Weighting Factor
FIG. 7. Periodic Impulsive Horizontal Wind Pre..ure
TABLE 1. Relation between Control Weighting-Factor Coeffi-
sequently, the effect of load combination is considered insig- cient and Maximum Force Exerted by Controllers
nificant for placement of controllers in bridge structures.
r Maximum force (kN)
In all the examples the state variables weighting factor q is (1 ) (2)
chosen equal to one. That is, all state variables are given equal
weights. However, different values are chosen for the control 1.0 1.29
0.5 1.92
weighting factor r in order to investigate the effect of changing 0.1 3.81
the level of control forces on the response of the open-loop
structure. There is an inverse relationship between this factor
and the level of the force exerted by the actuator as noted in
(4).

Example 1
This is a single-span truss bridge with a span of 48.6 m, a -Open loop
height of 6 m, and a width of 6 m (Fig. 3). The structure has """SchemeA
- - - SchemeB
292 members and 76 nodes. This results in 432 state variables ~-ScherneC
(three displacements and three velocities for each node of the 5 -0.5

structure). Schemes A, B, and C for placement of the con-


trollers are identified in Fig. 3. Three different values are cho-
sen for the control weighting factor r, starting with r 1.0 = !... -I

-1.5

and then decreasing it to 0.5 and 0.1. i


Example 2
FIG. 9. Vertical Displacement at Midpoint of Top Chord of
This is a two-span truss bridge with a total span length of Tru.. of Example 1 Due to Traffic loading, Using Three Different
72 m, a height of 6 m, and a width of 6 m (Fig. 4). The Schemes A, B, and C for Placement of Controllers

222/ JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997

J. Struct. Eng., 1997, 123(2): 218-226


example 1 under the traffic loading using scheme A. In the
rest of the paper a value of r = 0.1 is used for the control
weighting factor.
Using the three different schemes A, B, and C for placement
of the controllers, the vertical displacement at the midpoint of
-Open loop the top chord of the truss of example 1 due to traffic loading
..•. "Scheme A is presented in Fig. 9. Figs. 10 and 11 present similar results
- - - SchemeB
-SchemeC
for the horizontal displacement at the same point due to wind
loading and earthquake loading, respectively. It is observed
50 that schemes A and B reduce the response of the structure
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

Time (sec.)
more effectively than scheme C, and scheme A reduces the
response only slightly more than scheme B. For the case of
wind loading the maximum response in the controlled struc-
ture is reduced by 52, 48, and 40% using schemes A, B, and
-3

2
FIG. 10. Horizontal Displacement at Midpoint of Top Chord of
'n'U88 of Example 1 Due to Wind Loading, Using Three Different
Schemes A, B, and C for Placement of Controllers
~
ij
.. 1.5
1
B
;;
.
B
30 os
:5
t
._~._-

~
;; 20 -Open loop
os -Open loop E .... - -Scheme A
---···SchemeA .;i
:;-
0.5 - - - Scheme B
- - - SchemeB ----- Scheme C
;
-SchemeC

i
i
50
.~

"~ Time (sec.)

= -0.5
·e
~
-I

FIG. 13. Horizontal Displacement at Midpoint of Top Chord of


Each Span of 'IWo-Span Bridge of Example 2 Due to Wind Load-
Ing
FIG. 11. Horizontal Displacement at Midpoint of Top Chord of
'n'U88 of Example 1 Due to Earthquake Loading, Using Three 20
Different Schemes A, B, and C for Placement of Controllers

1.5

! -Open~oop
1 ..... -Scheme A
- - - SchemeS
t 0.5 --_.._- Scheme C

.s
i 0 4--ft+P"-fo.<::d.......t---f+--\---++----\+-----f-+--\--f---'lr< -Open loop
······SChemeA
] -0.5 - - - Scheme B
-5 .- --- Scheme C
• -I

J.. -1.5

:S
"! -2 -20

~ -2.5
FIG. 14. Horizontal Displacement at Midpoint of Top Chord of
Each Span of Two-Span Bridge of Example 2 Due to Earthquake
FIG. 12. Vertical Displacement at Midpoint of Top Chord of Loading
Each Span of 'IWo-Span Bridge of Example 2 Due to 'n'afflc
Loading
t 2.5

~ 2
Fig. 8 displays the vertical displacement at the midpoint of { 1.5
the top chord of the truss of example I under the traffic load-
1iO~
ing using scheme A for both the open-loop system (uncon-
trolled structure) and the closed-loop system (controlled struc-
ture), using three different values for the control weighting
oi-
l; 1 0 [3
- - - H-2m
······H-6m
-0

1
11.0.5

factor r. The maximum displacement in the controlled struc- -J


-1.5
ture is reduced by 30, 37, and 50% for r values of 1.0, 0.5,
and 0.1, respectively. It is observed that with a proper control J ·2

weighting factor, it is possible to reduce the maximum re-


I -2.5
;l>

sponse of the controlled structure under traffic loading to 50%


FIG. 15. Vertical Displacement at Midpoint of Top Chord of
of the maximum response of the uncontrolled structure. Table Curved n-uss of Example 3 under 'n'afflc Loading for Open-Loop
I shows the relation between the control weighting-factor co- Systam (Uncontrolled Structure) for Three Different Values of
efficient and the maximum force exerted by controllers for Height H, Shown In Figure 5: 0, 2, and 6 m

JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997/223

J. Struct. Eng., 1997, 123(2): 218-226


'. using schemes A, B, and C, respectively. Thus, another con-
""
'.
'. ., clusion is that controllers can reduce the response of a contin-
'.
. ,,, ..', .,
, ,
,
uous multispan bridge more effectively than that of a single-
,
. span bridge.
Fig. 15 displays the vertical displacement at the midpoint
"
~ of the top chord of the curved truss of example 3 under the
~ traffic loading for the open-loop system (uncontrolled struc-
ture) for three different values of height H shown in Fig. 5:
0, 2, and 6 m. It is interesting to note as height H increases,
the need for control forces in the vertical direction is reduced.
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

~.
~ : Fig. 16 displays the horizontal displacement at the midpoint
','
of the top chord of the curved truss of example 3 under the
-4
wind loading for the open-loop system (uncontrolled structure)
FIG. 16. Horizontal Displacement at Midpoint of Top Chord of for two different values of height H: 0 and 6 m. It is observed
Curved TrU88 of Example 3 Due to Wind Loading for Open-Loop that the horizontal displacement increases with the higher
System (Uncontrolled Structure) for Two Different Values of value of H, thus the greater need for controllers in the hori-
Height HShown In Figure 5: 0 and 6 m zontal direction. Using H = 6 m, Figs. 17 and 18 present re-
sults for the horizontal displacement at the midpoint of the top
chord of the curved truss of example 3 due to wind loading
and earthquake loading. It is observed that schemes A and B
reduce the response of the structure more effectively than
scheme C, and scheme A reduces the response only slightly
more than scheme B. For the case of wind loading the maxi-

~
-----

-Open loop
. - '" -Scheme A mum response is reduced by 46, 42, and 32% using schemes
- - - Scheme B
--", --- Scheme C
A, B, and C. For the case of earthquake loading the maximum
---_.'-----
response is reduced by 67, 62, and 49% using schemes A, B,
70 80
andC.
Time (sec.)

CONCLUSIONS

In this research a computational model has been developed


FIG. 17. Horizontal Displacement at MidpoInt of Top Chord of
Curved Tru88 of Example 3 Due to Wind LoadIng UsIng Three for optimal active control of large structures subjected to dy-
Different Schemes A, B, and C for Placement of Controllers namic loading. The computational model has been applied to
three different types of bridges subjected to impulsive traffic,
impulsive periodic wind, and dynamic earthquake loadings.
-Open 1;;;P] Three different schemes have been investigated for placement
-""'SchemeAI of controllers. It is demonstrated that through the use of con-
-. - - Scheme Bj' trollers the response of a bridge structure to dynamic loadings
-'-SchemeC
can be reduced substantially. The magnitude of the reduction
is the largest for the case of dynamic earthquake loading.
On the question of optimal placement of controllers, the
Time (sec.) following conclusions are reached and recommendations
made.
For bridge structures controllers are recommended to be col-
located along the members lying in the cross-sectional planes
and perpendicular to the vertical plane of bridge structure,
which includes the principal modes of vibrations.
FIG. 18. Horizontal Displacement at Midpoint of Top Chord of For different types of bridge structures schemes A and B
Curved Truss of Example 3 Due to Earthquake Loading, Using reduce the response of the bridge structure more effectively
Three Different Schemes A, B, and C for Placement of Control- than scheme C, and scheme A is only slightly more effective
lers than scheme B. Since the number of controllers in scheme B
is roughly one half of that in scheme A, scheme B is recom-
C, respectively. For the case of earthquake loading the maxi- mended as the most economical solution for placement of con-
mum response in the controlled structure is reduced by 73, 68, trollers.
and 55% using schemes A, B, and C, respectively. Compared with simply supported bridges, the response of
Fig. 12 shows the vertical displacement at the midpoint of continuous bridge structures can be reduced more effectively
the top chord of each span for the two-span bridge of example by controllers. In other words, the required level of actuator
2 due to traffic loading. Figs. 13 and 14 show the horizontal forces for continuous bridges is less than that for simply sup-
displacement at the midpoint of the top chord of each span for ported bridges.
example 2 due to wind loading and earthquake loading. The The curvature of a curved bridge reduces the response due
horizontal wind loading is applied to one half of the bridge to the traffic loading in the vertical direction (plane of the
only to obtain the most critical response. Again, it is observed curvature), thus reducing the need for controllers to reduce the
that schemes A and B reduce the response of the structure response in the vertical direction. However, the curvature in-
more effectively than scheme C, and scheme A reduces the creases the level of actuator forces needed to reduce the re-
response only slightly more than scheme B. For the case of sponse due to wind and earthquake loadings in the horizontal
wind loading the maximum response is reduced by 54, 50, and direction.
42% using schemes A, B, and C. For the case of earthquake This research is currently be extended to multistory high-
loading the maximum response is reduced by 75, 71, and 61% rise building structures.
224/ JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997

J. Struct. Eng., 1997, 123(2): 218-226


ACKNOWLEDGMENTS n =eE>V T
(microtasking and vectorization)
This research is based partly on a research project sponsored by Cray
Research Inc. Computing time was provided by the Ohio Supercomputer
5. For p = I until n p do (microtasking)
Center, Mendota Heights, Minn.
j=p
APPENDIX I. PARALLEL-VECTOR ALGORITHM FOR 6. For k = 1 until 2N do (vectorization)
RESPONSE OF CLOSED-LOOP SYSTEM
(The same algorithm can be used for theJesponse of the
open-loop system of equations by re£lacing A by A.) Next k
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

Input: n, N, Nt., np , 8T, T, A, B o , X( >, f~o>, e, and V. ~Soluti~n for k = 1 until Nt. do (vectorization)
of the complex eigenvalue problem of the unsymmetnc matnx
A is obtained per Saleh and Adeli (1996)]
m = 0; i =2 Next k
for j = 1 until 2N do (vectorization)
j=j + np
xt) =XJO)
if j ::=; 2N then go to 6.
Nextj Nextp
for p = 1 until np do (microtasking) if (m + 1 $ T/8T) then
for j = 1 until 2N do (vectorization)
j=p

1. For k = 1 until 2N do (vectorization)


Nextj
[,(1)1.k =I j •k

Next k m=m + l,got05.

j =j + np Stop
if j ::=; 2N then go to 1.
Nextp APPENDIX II. REFERENCES
2. For p = 1 until np do (microtasking) Adell, H. (1992a). Supercomputers in engineering analysis. Marcel Dek-
ker, Inc., New York, N.Y.
j=p
Adell, H. (1992b). Parallel processing the computational mechanics.
3. For k = 1 until 2N do Marcel Dekker, Inc., New York, N.Y.
American Association of State Highway and Transportation Officials
(AASHTO). (1993). Standard specifications for highway bridges, 15th
8TAj •,
[, 1.k =Ij •k =
(I) (I_I) • •
+ n + 1 -,. [, ]I.b 1 I, 2, ... , 2N (vectonzation) Ed., Washington, D.C.
American Institute of Steel Construction (AISC). (1989). Manual of steel
construction-allowable stress design, 8th Ed., Chicago, III.
Next k Arabel, A. (1981). "Controllability measures and actuator placement in
oscillatory systems." Int. J. Control, 33(3), 565-574.
j=j + np Casti, J. L. (1987). Linear dynamical systems. Academic Press, New
York, N.Y.
if j ::=; 2N then go to 1. Chopra, A. K. (1995). Dynamics of structures: theory and applications
Nextp to earthquake engineering. Prentice-Hall, Inc., Englewood Cliffs, N.J.
Dyke, S. J., Spencer, B. F. Jr., Quast, P., Kaspari, D. C. Jr., and Sain, M.
i=i +1 K. (1996). "Implementation of an active mass driver using acceleration
feedback control." Microcomputers in Civ. Engrg., 1I(5), 305-323.
if i ::=; n - 1 then go to 2. Furuya, H., and Haftka, R. T. (1995). "Placing actuators on space struc-
For p = 1 until np do (microtasking) tures by genetic algorithms and effectiveness indices." Struct. Opti-
mization, 9(2), 69-75.
j=p Housner, G. W., Soong, T. T., and Marsi, S. F. (1996). "Second gener-
ation of the active structural control in civil engineering." Microcom-
4. For k = 1 until 2N do puters in Civ. Engrg., 1I(5), 289-296.
Khot, N. S. (1994). "Optimization of controlled structures," Advances
E>j.k =I j •k + 8TAjJ [,<n-l)hb 1= 1,2, ... , 2N(vectorization) in design optimization, H. Adeli, ed., Chapman and Hall, London, Eng-
land,266-296.
Next k Kobori, T. (1996). "Future direction on research and development of
for k = 1 until 2N do (vectorization) seismic-response-controlled structure." Microcomputers in Civ. Engrg.,
6T
1I(5), 297-304.

1
tJ.J•k = 0 E>J.k dt
Marks II, R. J. (1993). Advanced topics in Shannon sampling and inter-
polation theory. Springer-Verlag, Berlin, Germany.
Meirovitch, L. (1985). Introduction to dynamics and control. John Wiley
& Sons, Inc., New York, N.Y.
Next k Oz, H., Farag, K. A., and Venkayya, V. B. (1990). "Efficiency of struc-
ture-control systems." J. Guidance, Control, and Dyn., 13(3), 722-
j=j + np 735.
Sakamoto, M., Sasaki, K., and Kobori, T. (1992). "Active structural re-
if j ::=; 2N then go to 4. sponse control system." Mechatronics, 2(5), 503-519.
Nextp Saleh, A., and Adeli, H. (1994a). "Microtasking, macrotasking, and au-
totasking for structural optimization." J. Aerosp. Engrg., ASCE, 7(2),
r = etJ.VTBo (microtasking and vectorization) 156-174.
JOURNAL OF STRUCTURAL ENGINEERING / FEBRUARY 1997/225

J. Struct. Eng., 1997, 123(2): 218-226


Saleh, A., and Adell, H. (1994b). "Parallel algorithms for integrated struc- B = input matrix;
tural/control optimization." J. Aerosp. Engrg., ASCE, 7(3), 297-314. Bo = applied dynamic-loading matrix;
Saleh, A., and Adell, H. (1996). "Parallel eigenvalue algorithms for large e = right-hand eigenvector matrix;
scale control-optimization problems." J. Aerosp. Engrg., ASCE, 9(3), f = control input vector;
70-79.
f o = vector of applied dynamic loading;
Schulz, G., and Heimbold, G. (1983). "Dislocated actuator/sensor posi-
tioning and feedback design for flexible structures." J. Guidance. Con- G = control optimum-gain matrix;
trol. and Dyn., 6(5), 361-367. I = identity matrix;
Stubbs, N., and Park, S. (1996). "Optimal sensor placement for mode M = structure mass matrix;
shapes via Shannon's sampling theorem." Microcomputers in Civ. N = number of degrees of freedom;
Engrg., 11(6),411-419. Nt = number of actuators;
Subramaniam, R. S., Reinhorn, A. M., Riley, M. A., and Nagarajaiah, S. Nt. = number of dynamic loads applied at predetermined de-
Downloaded from ascelibrary.org by UFRGS - Universidade Federal do Rio Grande do Sul on 10/25/21. Copyright ASCE. For personal use only; all rights reserved.

(1996). "Hybrid control of structures using fuzzy logic." Microcom- grees of freedom;
puters in Civ. Engrg., 11(1),1-17. np = number of processors;
Yang, J. N., Li, Z., Danielians, A., and Liu, S. C. (1992). "A seismic P = Riccati matrix;
hybrid control of nonlinear and hysteretic structures-I." Engrg. Q = state weighting matrix;
Mech., ASCE, 118(7), 1423-1440.
q = wind pressure;
R = control weighting matrix;
APPENDIX III. NOTATION T = total time;
ug = vector of ground acceleration;
The following symbols are used in this paper:
V = left eigenvector matrix;
X = state variable vector (displacements and velocities);
=
X time derivative of state variable vector;
A = plant matrix; 8T = time increment; and
A = closed-loop matrix; 8(t) = delta function.

226/ JOURNAL OF STRUCTURAL ENGINEERING I FEBRUARY 1997

J. Struct. Eng., 1997, 123(2): 218-226

You might also like