Professional Documents
Culture Documents
Abstract
1
Flexure tUbe
Coil 8 Armature
I
Table 1: System Constants
Constant Description
A, = 3.4238 in2 Surface area of piston
m = 1.1775 Mass of piston and attached
platform
= 100 9 Damping coefficient
Fsc = 5.0 lb Seal friction force
y = 10000 Seal function constant
w = 0.406 in Width of orifice opening
Cd = 0.61 Fluid flow coefficient
p s = 1000 % Supply pressure
lbZn Return pressure
pr = 25
- *, p = 0.000078
p = 200,000 3
Density of hydraulic fluid
Bulk modulus of hydraulic fluid
Figure 2: Servovalve Structure Re = 101 R Resistance of servovalve coils
Le = 0.05 H Inductance of servovalve coils
2. Mathematical Model 01 = 44 Modeled resistance in spool
movement
Let XI,x2, and x3 be the displacement, velocity and ff2 = 0.8 am;n,ec Modeled inductance in spool
acceleration of the working piston, then we have the movement
natural relationship $1 = 2 2 and x 2 = 2 3 . Based on kl = 3.12307 Input gain
three assumptions [3, p.161, the relation of the pres- IC2 = 5.34045 Feedback gain
sures between left and right chambers of the working
piston is governed by l j , = -lj,(Vo - A,xl)/(K + Centro Vertical, as desired reference acceleration can-
+ +
A , q ) or p, ps = P, P,, where V, is the initial didates. The first two acceleration data has a band-
chamber volume with the working piston at the center width around 20 Hz, and the third acceleration data
position. Define the load pressure Ap = pa -ps; then have a bandwidth of about 40 Hz. To unify and re-
the relationship between the acceleration and the load duce the time intervals of the earthquake accelera-
pressure is given by 2 3 = (A,Ap - <zZ)/m.By elimi- tion data, a cubic spline interpretation is applied to
nating the load pressure Ap, the nonlinear model for the three earthquake acceleration data so that a more
the shaker table has three more equations and one smooth data could be used to further integration and
switching surface, with zero initial conditions. The differentiation. A scaling of both the acceleration
+
first two additional expressions are for ks $ 2 3 , for magnitude and the time is made such that the three
the cases x 5 2 0, and 2 5 < 0; earthquake acceleration data and the corresponding
reference signals are compatible for the practical uses.
Let Ct and C, be the scaling factors of the time and
the acceleration magnitude, respectively. Then the
scaled acceleration data is given by t,,, = C t t o l d and
a n e w = C m a o l d , where sold is the magnitude of the
original acceleration data with respect to the original
time t o l d , and anew is the magnitude of scaled accel-
eration data with respect to scaled time t,,,. The
scaled and smoothed three earthquake acceleration
histories are shown in Figure 3.
1 01
x5 = -24 - -x5
0 2 0 2
2
(a) 1934 El Centro South
I
1 0 1z
TI--. <s-oo"d-,
3
signals
Q4,O = [I1 02." 136 037 ' * * 156 057 " * 0701 j
(14)
where l i or O i represents the i-th element of the ma-
+
trix, and Qi,j = 0, V ( i j ) = 4, i , j = 0,1, * . * , 4 ,
W I except Q 4 0 given above, such that Q40x: = xCsl +
x& + x&. The differences among the simulation re-
1 0 12
TIm-- <S-PPnd*>
(a) 1934 El Centro South sponses of linear and 3rd-order controls are not signif-
icant both in time and frequency domains. It can be
seen from the simulation results that the open-loop
response corresponding to reference command signal
U,, computed by inverse method, decreases the stan-
dard deviation of acceleration tracking error about
95% of the standard deviation of the reference state
z,, for both 1934 El Centro South and 1934 El Centro
West cases, 86.87% for the case of 1940 El Centro Vcr-
tical. The linear optimal control decreases the stan-
(b) 1934 El Centro West dard deviation of acceleration tracking error about
30% of the standard deviation of open-loop acceler-
ation tracking error in three cases. The 3rd-order
optimal control decreases the standard deviation of
acceleration tracking error about 2% of the standard
deviation of linear optimal control, for both 1934 El
Centro South and 1934 El Centro West cases, and
only 0.35% for the case of 1940 El Centro Vertical.
With a paired reference command signal { u r , x r } ,
both linear and nonlinear tracking controls for the
shaker table have less tracking errors, comparing with
tracking errors in the case of mismatching between
the elements of command signal, i.e. only one refer-
Figure 5 : Reference Command Signals ence command signal z, being used. To demonstrate
where Qif denotes the symmetric tensor inverse of the advantage of nonlinear controller over linear con-
Q0.2,and is defined by troller for the tracking control of the shaker table, set
=
u r ( t ) 0 for all t 2 0, so that closed-loop nonlinear
k i system for the shaker table has a potential to move
outside the linear region. The time responses of the
acceleration tracking errors for the case of mismatch-
ing between of the elements of the command signal
with U, 0 are shown in Figure 6. The variances
5 . Controller Simulation for desired acceleration (1934 El Centro West), cor-
responding linear and nonlinear acceleration tracking
The nonlinear cost functional is chosen in the form errors are 14.2593, 1.0283 and 0.0294, respectively, as
J= 1 03
4
MeasuredAccelerations of Shake Table (elc4Ov)
150
I
c-
100 i .
p 50
8
e o
8 -=
-4 E -50
-100
I ,
2 4 6 6 io 12 14
Time (sec.)
(a) Linear Control
20
r8
e o
5
-20
2.b5 2:1 2.;5 2:2 2.;5 213 2.k 2:4 2.45 215
I , I
10 1-
Time (sec.)
TI,”. <.-oo”dr,
Figure 6: Acceleration Tracking Errors (1934 El Centro Engineering at the University of Notre Dame, and in
West, U? E 0) part by the Department of Electrical Engineering at
the University of Notre Dame.
the polynomialized model of the shaker table with a
paired reference command signal. For the case of mis- References
matching between the elements of the desired com- [l] B. F. Spencer, M. K. Sain, S. J. Dyke, and
mand signal, i.e. ur(t)G 0 for all t 1: 0, a practical P. Quast, “Acceleration Feedback Control Strategies
state estimator has t o be designed and implemented for Aseismic Protector,” in Proceedings of the 1993
so that we can demonstrate the advantage of nonlin- American Control Conference, pp. 1317-1321, June
ear control over linear control experimentally. There 1993.
are two desired acceleration signals are used for the
experiments of the tracking control of the shaker ta- [2] J. A. O’Sullivan, Nonlinear Optimal Regulation
by Polynomic Approximation Methods. Ph.D. Disser-
ble: 1934 El Centro West and 1940 El Centro Verti-
tation, Department of Electrical and Computer Sci-
cal. Later has a wider frequency range for the desired
ence Engineering, University of Notre Dame, May
acceleration spectrum. The acceleration tracking re-
1986.
sponses and acceleration tracking errors are shown
in Figure 7. Two solid lines are desired and open- [3] D. P. Newell, “Modeling and Control of a Non-
loop acceleration responses of the shaker table. There linear, Hydraulic Based Seismic Simulator,” M.S.
is some time shifting for measured open-loop, linear Thesis, Department of Electrical Engineering, Uni-
and nonlinear controlled acceleration responses of the versity of Notre Dame, November 1993.
shaker table due to sampling trig error when using the [4] M. Vidyasagar, Nonlinear Systems Analysis.
desired displacement signal as measurement starting Englewood Cliffs, New Jersey: Prentice Hall, sec-
trig for each conducted experiment with respect to ond ed., 1993.
same desired acceleration signal. The standard devi-
ations of acceleration tracking errors for open-loop, [5] Hongliang Dai, Investigations o n the Control
of a Hydraulic Shaker Table and of Magnetorheologi-
linear and nonlinear control are 6.2057, 7.3188 and cal Dampers f o r Seismic Response Reduction. Ph.D.
7.4908, respectively, corresponding to the standard Dissertation, Department of Electrical and Computer
deviations for the measured desired acceleration sig- Science Engineering, University of Notre Dame, Au-
nals are 8.0190, 8.0187 and 8.0187. gust 1996.
7. Acknowledgements