2007 IEEE International Conference on Control and Automation WeC3-5

Guangzhou, CHINA - May 30 to June 1, 2007

Three State Controller Design of Shaking Table in

Active Structural Control System
Xu Yang, Hua Hongxing Han Junwei
State Key Laboratory of Mechanical System and Vibration IEST
Shanghai Jiao Tong University Harbin Institute of Technology
Shanghai, China Harbin, China
xuxuyangyanggsj tu.edu.cn [email protected]

Abstract-The purpose of this paper is to develop an approach to In this paper, firstly, the active control experimental system
improve the tracking characteristics of a small-scale shaking is introduced, and then the dynamic model of electro-hydraulic
table actuation system, which is used in active structural control shaking table actuation system has been constructed and
system. In this paper, firstly, the active control experimental analyzed. Based on the constructed model, three states
system is described in detail. Then, the process of constructing controller of shaking table is designed to extend the
analytical model of servo-hydraulic actuation system is presented. acceleration bandwidth and improve system damping. At last,
The proposed model, in the form of the total shaking table the effectiveness of controllers is experimentally validated. The
transfer function, accounts for the specific physical results indicated that the shake table is operating with desired
characteristics of the small-scale shaking table. Due to low levels of performance.
natural frequency and small hydraulic actuator damping, the
small-scale shaking table cannot meet the experimental II. EXPERIMENTAL SYSTEM SETUP
requirement. Thus, three states feedback and three state feed-
forward control algorithm based on pole-assignment principle is Active structural control system mainly consisted of
applied for extending the acceleration bandwidth and improving shaking table, hydraulic actuator, a three-floor test structure,
system damping. Lastly, the shaking table experiments are tested and signal modulator and sensors and so on, as shown in Fig.
and the experimental results indicate that the controller is 1.
effective. The dimension of aluminum platform of shaking table is
Keywords-strucutral active control; shaking table; three state 0.9mx 0.7 mx 0.026 m and weight is 44.3Kg. The motion of
controller; modeling; pole-assignment the table is constrained to a single horizontal hydraulic
cylinder. The actuator has an effective piston area of
I. INTRODUCTION 294.5 x 10-6m2 and ± 0.023m dynamic stroke. Hydraulic flow
Active structural control has made tremendous progress of the actuator applied in experimental system is controlled by
over the last two decades and has good control effect on many a symmetric servo-valve, which have OOHz of hydraulic
applications. In active control experimental system, shaking natural frequency. The maximum force which servo-actuated
table is one of the most important equipments. Due to high hydraulic cylinder can provide is 4.7kN.
building cost, availability of the large-scale shake tables is The test structure mounted on the shaking table is a three-
limited for testing. Thus, in order to validate proposed designs floor building with active mass damper. The frame of test
or new algorithms, many investigators are devoted to small- structure is constructed of steel, and with a span of O.5m. The
scale shake tables [1]. To ensure the results validity on small- test structure is 1.58m in height, and has a weight of 304.3kg,
scale shake tables, the fidelity of the shaking table should be distributed evenly among the three floors. The first three
paid more attention. The focus of this paper was to develop a modes of test structure, obtained by Finite Element software
high-performance, low-cost control method to enhance the (ANSYS), are at 5.38Hz, 15.68Hz and 22.53Hz respectively
tracking performance of the actuator. [5].
Generally, shaking table should have a certain acceleration The controller used in this experiment is implemented on
response bandwidth and enough stability margins. In our digital computer. The DSP board employed ACL 8312 have
experimental system, the small-scale shaking table cannot 12-bit differential A/D and D/A converters locally on the board.
meet such requirement due to low natural frequency and small A variety of sensors and signal modulator are assembly to
hydraulic actuator damping. Thus, to obtain the desired monitor the state of the hydraulics and the motion of the
acceleration response bandwidth and stability margins without platform. Real-time workshop of rapid prototype technology in
Mathworks is applied to control the experimental system.
changing original shaking table structure, an effective control
method must be developed in our experimental system. Three III. HYDRAULIC ACTUATOR MODELING
states control strategy is one of the most effective methods, The challenge of developing a control design is to establish
which can reassign closed-loop system poles with pole- relationship between the input of shake table and its output
assignment theory. This method can extend the acceleration response. To provide a rational basis for the design of three
response bandwidth of shaking table and improve system
stability [2-4].

88

1-4244-0818-0/07/$20.00 (© 2007 IEEE

 yV Y, Y2 Y3

I Al L ki k

acceleration p-
X W P2m mF0 ~~~X~
mi
~W | -3t Fd
sensor 4 Ql Q2
acceleration
sensor 3 test structue|
Ps I P.
acceleration
sensor 2 srv alve|
Fig. 2 Schematic diagram of equivalent load of hydraulic actuator system
acceleration
sensor 1 hydraulic cylinder Where, Vf is total volume of the actuator two chambers, /3e
displacement is effective bulk modulus of the fluid, Ctc is the total leakage
sensor shaking table
coefficient.
D/A 3) Force balance equation
CController output d2
AA Y
AC:L83j12 signal modulator
= APL
FT g L m
~~~dt2 +k (yP - y ) + Fd (3)
DSP acqusitiOn Sensor output 1
board
Real-Time Work Simulation
Control Program dt2
Fig. 1 Experimental system setup
states controller, in this section, the system model based on the k2 (y1 y2) k3(y2 y3) m2 dt2
d2 ()
physics of the process is constructed and analyzed [6].
For some shaking tables, the inertial load is the primary d 2y
k3(Y2 -y3) =m3 dt2 (6)
load. Elastic load is usually very small and can be ignored. But
in our experimental system, actuator system attached test Performing Laplace transformation to (1) - (6), the transfer
structure is equal to the multi-degree-of-freedom elastic function from actuator piston displacement yP to flow rate Q
resonant load, which takes up a large scale in all loads system.
Thus, elastic load must be taken into account and cannot be can be expressed as
ignored. (1/A)*(b6s6 +b4s4 +b2s2 +bo) (7)
Fig.2 shows schematic diagram of equivalent load of a9s +a8s8 +a7s7 +a6s6 +a5s5 +a4s4 +a3s +a2s2 +a1s
loaded hydraulic actuator system. Where, the symbol p, As the transfer function above is very complicated, factors
represents supply pressure, p0 is return pressure, i is servo- of transfer function are directly listed as follows.
valve input electric current, A (i = 1,2) represents the
a9 4A2 mmJn2m3 8)
effective cross-sectional area of actuator piston, because of
symmetry, here A1 =A2 Fd is external load force acted on Kace mmIm2m3 (9)
hydraulic actuator, m (i 1,2,3) is the mass of the i th floor, a, =A2
yi (i = 1,2,3) is the displacement of the i th floor relative to a [L 2(mmAM2k3+mNm3k2+mm1M3k3+mm2m3k1 (10)
ground, k, (i =1,2,3) is the structural stiffness of the i th
+mm2m3k2 +mlm2m3kl) +mm2m3]
floor, yp is the horizontal displacement of piston.
According to the hydraulic actuator three foundational a6 =[ A2 (mnyM2k3 +mnM3k2 +mNm3k3 +mm2m3k (11)
equations, the dynamics characteristic of hydraulic actuator +mm2m3k2 +±nm2m3k )]
system are given by
1) Valve flow equation V
a5 =[t,2 (mm1m2k3 +mm2k1k3 +mmft1& +mm3k1k3
QL KqXv KCPL (1) 4pe,4
+mmk2k3 +mm3kk3+nimAkAk3 +mnM3k1k +±nm3klk3
~~~~~~~~~(12)
Where, QL is flow rate, Xv is servo-valve input, Kq is flow
+m2m3k1k2)+(m1m2k3 +m1m3k +±nm3k3
gain, Kc is flow-pressure coefficient of servo-valve, PL is +m2m3k1 +m2m3k2)]
load pressure drop.
2) Continuity equation of actuator a4 =A 2 (mm2k k3 +mm2k k3 ±mm3k k 2 mm3k1k3
dy V dP +mm2k3k2 +mm3k2k3+±mm2k2k3+±mm3k1k2 (13)
QL=A dt~±C P± Vtt (2) +mlm3k1k3 +m2m3kk2 )]
t,L4,8,dt

89
 O S)
VI
a3 =[ 2
+
4JJleA (m+m, +m2 +m3)k,k2k3 +(m,k2k3 +m2kik3
mk1k2 + mk1k +mk3k2 +m3k2k3)]
a2 =L Kce (m+ml +m2 +m3)klk2k3

X
a, = k1k2k3
b6 = m1m2m3

K IV
W4 s6+hs4±+1+h)

j +sljcS +(± +as7

-
-4
- - - "D
(14)

b4 = (m1m2k3 + m1m3k2 + m1m3k3 + m2m3k1 + m2m3k2)

b2 = (mk2k3 +m2k1k3 +m3k1k2 +m3k1k3
+m2k2k3 +m3k2k3 )
bo = k1k2k3
Consider electro-hydraulic servo valve as a second-order
element, system open-loop transfer function can be combined
to yield

+±c 6 s6+±cs ±c4s4±c+±a(+ s2±+qs)

Where, Kvm =KaKqKf 1/ A is the system total open-loop
gain.
Through calculating, natural frequency cWhm of system
hydraulic actuator can be achieved, cohm =161.2 rad/s (about
25.6Hz). The expression of cohm is as following.

hmT (mm±+m2 +±m3)

(22)

Due to the natural frequency of servo-valve (about 100 Hz)

much higher than actuator natural frequency (about 25.6Hz),
transfer function of servo-valve can be simplified as
proportional element. Thus, the open-loop system transfer
function shown as (21) is further simplified as
C(s)- 8;,,js6±b~S4±b2sb0)
(15)
(16)
(17)
(18)
(19)
(20)

(21)

(23)
experimental system.

(L)

.0)
x
i

.1
rz
iM
0
100
---

-90

-180
10

1|)
3-1

-50 A-

-1 00 W.
--

-- -
-r

I I_
lllll
lllll
lllll
lllll
=rT+LiLi4
=

llllllll
-open-loop syasm
...

- I
-

----- - -

I-

10

System: G_cloop
Pole: -795
Damping: 1
Overshoot (%): 0
I
_+ -I + H H- - -I

r;.l- - - r -r-

-F

1-
lll lllll

-
-

10
r
.

_I

Frequency(Hz)
Fig. 3 System open and closed loop bode plot

System: G cloop
Pole: -99.6 + 45.3i
Damping: 0.91
Overshoot (%): 0.0999
Frequency (Hz): 1 7.4

Frequency (Hz): 127

Hl\o.,1| H
I_c1+Ose|-_1opls1y_+temH'-1l_%_1,j.*>-'11+FH1"-_'.l1', _l.1|"A
control method should be considered to introduce in our

lllll
lllll
lllll
lllll

(1' 1"111

lllll

iTnn-
HH-

1h&.||
,.,
1l fe
u LL
""'\
-
'''_
T
1

i.
_

\'
M

_'_ _'_
tT 2l
llllllll

-1- F Fl-ll

t.
llll
llll
llll
llll

1111
1111
1111

ll llll

,, ,1,,
llll
....... . .

11 ,,
fl

S
1111
1S1

lill
_1

f+_

_, -,- r r,l,T n - - n- n-rr.n r

lflil
,llllll
_F -'- L L il1 u _ _ J _ J _ e ,,,
''g'i"
t f I]

tit'' _ _

10
1_1 1111

*, ,,,,,
'_
___
_
'_'
1-t1X1
lllllll

r
llll
llll
llll
llll
-1- 1- 1- 1-1+ H - _ -I _ -I _ 1-1-l-1 1-

1111
1111

lllll

_ _ -1- _1-1-1-1d F
llll

........... ...

,lll
lil

10
C

(QS +±7s7
8 +a6s6 +a5s +a4s4 ±a3s3 ±a2s +±ls)

In order to conveniently analyze, an approximate form of -1 I.V

-600 -600-400 -200
1, 1.

0
transfer function can be obtained by using MATLAB program, Real axis
expressed as equation (24). Fig. 4 System closed loop pole distribution plot
By SIMULINK simulation, a frequency response
comparison between the system open-loop transfer function IV. THREE STATES CONTROL PRINCIPLE
and close-loop transfer function can be obtained, shows as the Three states feedback design is one of the most effective
Fig.3. The Fig.4 shows system close-loop zeros and poles methods to extend the acceleration bandwidth and improve
distribution. In Fig.4, circles present zeros and forks present system damping. It is based on the idea of pole assignment
poles. Analyzing Fig.3 and Fig.4, we note that there are three and can simultaneously control the system's displacement,
couples of zeros and poles near imaginary axis. Their peaks velocity and acceleration. Fig.5 shows foundational principle
are about at 5.22Hz, 15Hz, and 21.9Hz respectively, which is
correspond to first three modes of the structural system. This of three states control method. Where, Kdf K vf and K af
is consistent with results of ANSYS analysis. From Fig.4, we resent three states feedback control factors, K dr ' Kvr and
also know that poles -99 ± 45.3i are ones of determining
system bandwidth. It is clear that damping of close-loop Kar present three states feed-forward control factors.
system is about 0.9 and bandwidth is about 17Hz. The Adjusting acceleration feedback factor of three states feedback
bandwidth value is too low to meet experimental requirement may add system damping, adjusting velocity feedback factor
(bandwidth need at least up to 30Hz), so a new effective may extend system bandwidth but reduce system damping,
5.5x107 x(s2±1122Xs2± 9035Xs2 +19380) (24)
G(s)
( s(s+170.78)s+787.66 (S2 +50.59s+1716)s2 +32.39s+9123 s2 +0.09254s+19010

90
and adjusting displacement feedback factor may improve output signal d is
system gain. Therefore, desired system performance can be d Ku
achieved by rationally adjusting three states feedback factors. (,in)
'Vi
Three states feed-forward controller can further extent system R S2 +VfS + df s2
Whede(( 2+
d
24ss o 1)
bandwidth by canceling poles of determining system
bandwidth. Where, w0o d1f is original frequency of acceleration
displacement gainKdf control, 40 = Vf / 2d is damping of acceleration control.
Here, assumed Ku = 1, w0 = 0.4 Hz and 40 = 0.6.

V. CONTROLLER DESIGN
In this section, three states control strategy is introduced to
obtain system desired acceleration bandwidth and damping.
Because system damping under proportional control is up to
0.9, actuation system doesn't need to introduce acceleration
feedback to add system damping. Here, using velocity
feedback and displacement feedback is enough in controller
Fig.5 Principle of three states control
design. Fig.7 represents zeros and poles distribution of loaded
system along Kdf changing. When Kdf shown as Fig.7
Suppose to system transfer function G(s) as follows.
increasing, zeros and poles moved following arrowhead
G(s) _B(S)
(5 orientation. The arrowhead trend is that system natural
A(s) (25)
frequency improved and damping drops. From Fig.7, we
Expected transfer function Gm (s) is given by know that pole of determining close-loop bandwidth is a
couple of conjugated poles. If conjugated poles can be
Gm(s) mBi (s) (26) cancelled, the system frequency bandwidth will further
A,,,(s) improve.
Three states controller is to design S(s) and T(s) to make Let T(s) is as follow
T(s) = KarS + KvrS + Kdr (31)
the (27) existence.
B. (s) T(s)B(s) = (s + 99.6 + 44.5i)(s + 99.6 - 44.5i)
(27)
Am (s) A(s) + B(s)S(s)
Kar = 111981 (32)
Where S(s) is three states feedback polynomial, T(s) is
Kvr = 2 * 0.91/109(33)
three states feed-forward polynomial.
The system bandwidth should up to 127Hz. Here, T(s) is
S(s) and T(s) are given by second-order element and its purpose is to cancel a couple of
S(s)= Kd +Kv s+Ka s 2 (28) DU, , e x
T(s)= Kd +±Ks±+Ks2 (29) 0.982 0.962: Q¶925 .8 0.4 0A . ±

Design method above is based on position close-loop 100

1 92 I4 =S _, I=X t|i

system, but shaking table required acceleration control manner. ystem: ad8312_man_controller
Pole: -99.6 + 45.4i
Therefore, the three states filter shown as Fig.6 must be Damping: 0.91 >7
applied to transmitting the acceleration command signal R to s0 Overshoot (%). 0.102
position signal d. Frequency (Hz). 17.4 -ti-ox:
In Fig.6, the transfer function from target signal R to O.
700' 600' Soo: 400S 3000 200 t

acceleration System: ad831 2 man_controller

Pole: f793
150 - Damping: I
Overshoot (%): 0
Frequency (Hz): 127
-1 00 a9r*J92 I - r|t.t,

tL982P r 0b0.962W
99&9,5 G o0Mt 0o 4s
-'I Ju
-606 -700 -600 -500 -400 -300 -200 -100 0
Real

Fig. 6 Three states input filter Fig. 7 Zeros and poles distribution with three states feedback control

91
conjugation poles. Due to limitation of servo-valve, bandwidth response and actual actuator acceleration response. In Fig. 11,
is only up to 100 Hz. Fig.8 shows system zeros and poles dash line denotes desired acceleration signal and solid line
distribution plot after canceling conjugated poles. denotes actual actuator acceleration response. From Fig. 11 and
Combined position system with filter, acceleration frequency Fig. 12, it is clear that the two signal comparability is very well
response can be obtained shown in Fig.9. within a band-limited 1-30 Hz, and coh(e j) > 0.8. Due to
150n r I conditions limitation of the shaking table, actual actuator
O 0 .4- 0.E 0.,, 0 *7T-
.9

system: ad831 2man_controllerd

*9 * { 0 '
acceleration response is not totally identical equal to input
acceleration signal.
.QP2 'Zero: -99.2 + 45.2i
Damping: 0.91 I UM
Overshoot (%): 0 101
50
43gj Frequency (Hz): 17.3
mn 7

0 1
System:
Pole 795
_-' n ' i 5 --~ ~ O.
ad8Jl 2 mang controlleri 2E)T'
2B
2m ----------------

Damping I
-50 Overshoot (%)O0
Frequency (Hz): 127
-
5uf; L.

1 10 1 f; 2
20: 25; 30:
*1 nnn
-100 -
500 T 5 -I 1- r ;t
ile
2- 0.-i* 0.-~ g Me6 0.7 045 (D
S
-- - T - - - - - - - n - - - - - - - - - - - - - - -1- - - - - - - -r - - - - - - - r . . . . -1" ..
I!QU
-800 -700 -600 -500 -400 -300 -200 -100 0
Real a)a's ___
Fig. 8 System zeros and poles distribution with three states feed-forward
control -1 uuIU
5 10 15 20 25 30 35
inn1. Frequercy (Hz)
IT:
Fig. 10 Experimental result of acceleration bandwidth
A
0 °

1oJ -1 00 - j- l-: ,J.TT- -1-TI

-J
,,, nL
II T1. LI J.L I J. - rlG
-
U.rifL I- -1. rn - -1 -T
-J L

-200 0.6
.. E
-300 ..

180 I.r 1111111--91111111

z .nn n
11111111
TIT r rl n r
1111111
1 1l rn-*
1111111
1111l t 1Hl 1Hn 1Hn 1Hl
11111 {1111 11111 11111 111111
11111 11>_ 11111 11111 111111
'-
0 -- r n - rls TITI- - n - n n n T mn -2 . r T T n TIT - - r -1- r rls n r
1111 111 {'_\1 11111 1111
1- n -1-ls rln - n - T
111
1111 111 i 11111 1111 111
.\
4.-l 80
1111111 111111 1>11111
-- r n-rlnTITI- - n- n nnTnn - n - r T flnTIT - - r l-1- rrlnnr
1111111 111111
1- n-1-ln rln- n- T
-O.5
lllll llllll '1""' lllll llllll

-360 - r n - rls TITI- - n - n n n T nn -' n - r T T n t- t r

11111 111111 11111
n
11
r *1- n -1-ls rln - *n - T
sL 111111
1111111 111111 111111 S 1111111 1
-540 L
1111 111 11111 1111 111
llll lll ll lll llll lll
-,J.tL, -- - - -- I - -- I I I I -.1 - - I -.- ..- .-- .-- x

1-0 10 10 10 10 10 .zo
I ?rl
.zo ?.z ?R P, zoD.o
Time (s)
Frequency(Hz)
Fig. 9 System acceleration width Fig. 11 Comparison of expected random signal and measured random
signal
VI. EXPERIMENT
To demonstrate the effectiveness of three states controller, VII. CONCLUSION
we performed experimentally validation on the small-scale
shaking table. Inputting a white noise signal with the range of Several conclusions can be educed by analyzing dynamic
1-35Hz bandwidth to the actuation system, acceleration characteristic of shaking table actuator system.
system characteristic is obtained shown as Fig. 10. From 1) Three states control method based on pole-assignment
Fig.10, it is clear that bandwidth of loaded system is up to 30 made the performance of system can meet the active structure
Hz and peak values presenting at 5Hz, 15Hz and 22Hz experimental system requirements, which extent the
coincides with the natural frequency of structure. acceleration frequency band of shaking table actuator system
Coherence function is an important criterion to evaluate and improved the system stability without changing the
similarity of two random signals. In engineering, shaking table structure.
if coh(e jc) > 0.8, signal is thought as achieving reproduce. 2) The first three modes of the loaded shaking table
Fig. 11 shows a time comparison plot of actuator acceleration system obtained from the experimental results were coincident
response with the test structure and the desired acceleration. with the finite element analysis results. Consequently, the
Fig. 12 shows coherence function of desired acceleration finite element analysis method was proved feasible and can be
as a modeling tool.

92
 0.9 REFERENCES
J. Kuehn, D. Epp, and W.N.Patten, "High-fidelity control of a seismic
[1]
shake table," Earthquake engineering and structural dynamics, 1999, pp.
0 85 +-------------- = + - - ------
+ 1235-1254
[2] M. Tarokh, "Linearity of the Coefficients of the Closed-loop
68 '~~~~~~~~~~~~~~~~ ~~characteristic," INT1.
J. Control, vol. 45 no. 4, 1987, pp. 1383 1385
[3] J.W. Han, L. M. Yu, H. Zhao, and Z. Y. Wang. "Study of three state
controller of seismic simulating shaking table," Journal of Harbin
2 $, 'M, Institute of Technology, vol. 31 no. 3, 1999, pp. 21-24
.U675 [4] M. Tarokh. "On output feedback stabilization and pole assignment,"
lINT. J. Control, vol. 31 no. 2, 1980, pp. 399-408
o.7 , ±, ± -- -- [5] B. F. Spencer, S. J. Dyke, H. S. Deoskar. "Benchmark Problems in
Structural Control-Part I: Active Mass Driver System," Earthquake
Engineering and Structural Dynamics. vol. 27 no. 11, 1998, pp.1127-
0.66;5 L_____._____i_____i____i_____i_____i____ i___ 2247
0 5 10 15 20 25 30 [6] J. P. Conte, T. L. Trombetti, "Linear dynamic modeling of a uni-axial
Frequency (Hz) sevo-hydraulic shaking table system," Earthquake engineering and
Fig. 12 Coherence of expected signal and measured signal structural dynamics, vol. 29, 2000, pp.'13751404
3) When the elastic load of shaking table takes up a large
proportion of the total system, the elastic load cannot be
neglected. For loaded system, the two-order compensation is
effective to extent system bandwidth.

93