Beruflich Dokumente
Kultur Dokumente
3
and Josep M. Rossell
1
Institute of Information Theory and Automation, Academy of Sciences, Prague, Czech Republic
2
Department of Structural Mechanics, “Gh. Asachi” Technical University of Iasi, Romania
3
Department of Applied Mathematics III, Technical University of Catalunya, Barcelona, Spain
ABSTRACT
The paper is focused on overlapping decentralized control. This methodology was chosen to
be suitable for application to an international benchmark problem. After describing the
theoretical approach and the main steps in solving the overlapping control problem, the
paper presents the model of the bridge used for benchmark. A decomposition is proposed
and applied. Results are shown in terms of the responses to the external actions,
comparison with the non-controlled case and benchmark criteria. The methodology is
shown to be reliable and gives results close to or better than the centralized control.
1. INTRODUCTION
can be solved by standard decentralized methods. This design is made first by performing a
model reduction for each subsystem in expanded space. In this study the effectiveness of the
overlapping decentralized control approach is tested by numerical simulations. To do this,
the designed decentralized controllers are implemented into the original overall FEM
model. And, to measure the performance, closed-loop eigenvalue analysis, calculation of
evaluation criteria given in the benchmark problem, and analysis of time history response
for selected earthquake excitations are used.
2. PROBLEM STATEMENT
Consider the cable-stayed bridge illustrated in Figure 1. It is composed of two towers and
128 cables. The bridge is independently excited by three earthquakes longitudinal
acceleration: ElCentro NS 1940, Mexico City NS 1985 and Gebze NS 1999. These actions
have maximum values: 3.4170 m/s2, 1.4068 m/s2, and 2.5978 m/s2, respectively. Five
accelerometers and four displacement sensors are used to supply feedback information for
the control, which is produced by 24 hydraulic actuators located between the deck and the
towers and the end supports acting to apply longitudinal forces on the deck. A complete
physical description of the bridge, a finite element model and a MATLAB/SIMULINK
simulation framework are given in (Dyke et al 2000) as a benchmark for control design. A
centralized LQG control design is also presented in (Paulet-Crainiceanu et al. 2001).
3. SOLUTION
We propose the decomposition of the bridge model into two subsystems. Each subsystem
corresponds to one of the towers, the cables attached to it and the part of the deck where the
cables are attached. This is illustrated in Figure 1. Both subsystems are interconnected
BAKULE, PAULET AND RODELLAR 3
through the center of the bridge. It corresponds with the part of the deck where no cable is
attached. The overlapping decomposition procedure considers the towers strongly
connected via deck in the part where no cable is attached. When the original model is
extended, it defines a state space model in a larger space with the structure of disjoint
subsystems and interconnections.
More precisely, the overall original FEM model consists of 838 states. By properly re-
arranging the components of the state, input and output vectors, the overall model can be
split into two overlapping subsystems. These subsystems have 412 and 432 states. The
overlapping subsystem has 6 states. The overlapping common part includes one sensor but
no actuator.
subsystem 1 subsystem 2
First, the benchmark sample LQG design has been selected as a reference. Further, the
control subsystems have been defined with the same locations and models of sensors and
actuators as in the reference case. The global model has 8 control inputs and 13 measured
outputs. The expanded system has two subsystems with 412 and 432 states, 4 and 4 control
inputs, and 7 and 7 measured outputs, respectively. There are 14 measured outputs in the
expanded space because the overlapped part includes a sensor that is also expanded. It
means that the commonly shared states and the sensor are doubled in the expanded space to
obtain disjoint subsystems.
A decentralized control law is proposed for each free subsystem by combining its model
reduction and the LQG design on the reduced order subsystems. Model reduction first
forms a balanced realization and then condenses out the states with relatively small
controllability and observability grammians.
4 DECENTRALIZED CONTROL DESIGN FOR A CABLE-STAYED BRIDGE BENCHMARK
The following algorithm describes this overlapping LQG control design process.
3.3. Algorithm
1. Expand the original LQG problem with identified overlapping subsystems into a
larger expanded space.
2. Perform model reduction for each subsystem. Select a minimal order of the
subsystem’s states ensuring the stability of the reduced-order models.
3. Perform the LQG with preselected weighting matrices for reduced order subsystems.
4. Contract and implement the local controllers into the original overall FEM model and
run simulations.
5. Evaluate the results by computing the given benchmark evaluation criteria, the closed-
loop system eigenvalues and the dynamic responses, all in comparison to the
centralized control design reference case.
6. Tune the control laws by repeating the simulations for different weighting matrices
until acceptable results are reached.
For the overlapping decomposition of the FEM model described above, the model reduction
results in reduced-order stable subsystems with a minimal dimension of 34 states for each
subsystem. The decentralized LQG control design has been performed with the weighting
on the state defined by the identity matrix multiplied by a scalar q1. Some results are
summarized in the following.
x 10
4 TOWER SH.FORCE, GEBZE TOP TOWER DISPL. EL CENTRO
2 0.2
0.15 uncontrolled
1 controlled
0.1
Displacement(m)
Shear force (kN)
0 0.05
0
-1
-0.05
-2 uncontrolled -0.1
controlled
-0.15
-3
-0.2
0 10 20 30 40 0 10 20 30 40 50 60
Time(s) Time(s)
For one of the towers, Figures 2 and 3 show the base shear force and the top
displacement in comparison with the uncontrolled cases. The excitations are the
acceleration of Gebze and El Centro earthquake, respectively.
BAKULE, PAULET AND RODELLAR 5
From Figures 2 and 3, it is observed that the improvement achieved in reducing the
tower base shear force is obtained in exchange for an increase (within an acceptable range)
in the top tower displacement response with respect to the case without control.
In order to compare the results obtained with the overlapping decentralized control with
those given in the centralized reference case, Table 1 gives the first ten eigenvalues of the
closed-loop system modes. The overlapping decentralized case corresponds to the chosen
tuning parameters.
The first modes of vibration are changed by the proposed feedback controllers,
approaching the hinged (non-controlled) case. It can be observed that the overlapping
strategy is bringing the structure closer to the hinged case, keeping the advantage of a
reduction in tower base shear force (as the non-hinged structure would do).
1.5 200
1
1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8
trial no. trial no.
The benchmark evaluation criterion No. 6 (Dyke et al. 2000) is presented in Figure 4 for
the overlapping decentralized LQG control design during the tuning trials. Direct
comparisons with the benchmark sample centralized LQG control design case given by
6 DECENTRALIZED CONTROL DESIGN FOR A CABLE-STAYED BRIDGE BENCHMARK
Table 4 in (Dyke et al 2000) are included trough horizontal lines in all graphs. This
comparison is made for the three different earthquakes provided by the benchmark and
shows satisfactory fulfillment of the control objectives. Criterion No. 6 is the maximum
value of the displacement at the end of the bridge in the case of control over the same in the
case without (Dyke et al 2000).
Figure 5 shows the variation of one of the actuators maximum force during the tuning
process. The maximum forces in the four actuators groups were finally set from 932 to 934
kN. This setting was essential in establishing the LQG control parameters for each
subsystem.
4. CONCLUSIONS
The paper has presented simulation results of the overlapping decentralized LQG design for
the cable-stayed bridge benchmark performed in a SIMULINK/MATLAB scheme. The
main motivating ideas and objectives of the paper have been to begin to explore potential
usefulness of overlapping decentralization issues for active control of large structure as the
prototype bridge under study.
As effective measures for performance assessment and comparison, the benchmark
evaluation criteria and the analysis of time histories of selected bridge response variables
for different prototype earthquakes have been adopted. Additionally, the eigenvalues of a
number of significant modes have been checked for the closed loop obtained when
implementing the overlapping decentralized controller into the global finite element model.
The results look promising and confirm expectations. They are close to the sample
centralized case and lie within acceptable ranges. They satisfy also the requirements on
cable tensions that are known a-posteriori.
ACKNOWLEDGEMENTS
This work was supported in part by the ASCR under Grant A2075802, by the MEC under
Grant DGESIC-SB99 00787598, and by the CICYT under Project TAP99-1079-C03.
REFERENCES