Proceedings of the 2009 IEEE International Conference on Networking, Sensing and Control, Okayama, Japan, March 26-29, 2009

Vibration Suppression Control for a Multi-degree-of-freedom Structural System using an AMD with Restricted Stroke
Kazunobu Yoshida Faculty of Science and Engineering Shimane University Matsue, Shimane 690-8504, Japan
kyoshida@ecs.shimane-u.ac.jp

Itaru Matsumoto Department of Mechanical Engineering Yonago National College of Technology Yonago, Tottori 683-8502, Japan
i-matsum@yonago-k.ac.jp

Abstract The problem of suppressing the vibrations of a multi-degree-of-freedom structure system using an AMD (Active Mass Damper) with restricted stroke is considered. First the problem is reduced to the one with an input constraint by constructing a position servo system for the weight (the auxiliary mass) of the AMD. Then a saturating control is designed by a partial state feedback technique. The control law has the following properties: 1) a good control performance is obtained controlling the weight of the AMD within the stroke constraint; and 2) the control law is simple and easily implemented. The effectiveness of the control law is demonstrated by simulations.

I. I NTRODUCTION An active mass damper (AMD) is a mechanical device that suppresses the vibrations of a structural system using reaction forces generated by moving the auxiliary mass with an actuator connected between the structure and the auxiliary mass. Many applications of AMDs to real buildings and civil structures have already been reported. The number of applications to high-rise buildings has been over sixty since the Kyobashi Center building, the former Kyobashi Seiwa building, an 11-story building in Tokyo, Japan, was constructed by the Kajima Corporation in 1989; and these applications have been conducted mainly in Japan [1], [2]. The auxiliary mass of an AMD for a building structure is usually less than 0.4% of the total mass of the structure [3]; that is, a relatively small mass is to be moved with a limited amplitude to suppress the vibrations of the structure. Therefore, AMDs intrinsically dot not have an ability to absorb the energy due to large earthquake excitations, but they aim to suppress the vibrations of structures under strong winds or moderate earthquakes. Generally, the allowable amplitude of an auxiliary mass is severely restricted due to the size of the AMD or the installation space. In fact, the amplitude constraint of an auxiliary mass is one of the main reasons for the limited performance of AMDs. Thus, it is desired to develop a control law that effectively suppresses the vibrations of a structure system under the amplitude constraint of the auxiliary mass. Control laws for AMDs that have been in practical use are mainly state feedback controls with constant gain, obtained by LQ optimal control theory or H control theory [3]. These gains are designed to satisfy the amplitude constraint of the auxiliary mass for the maximum possible disturbance (wind or earthquake excitation), so they do not make the most of the performance of the AMD for ordinary disturbances.

For the problems considering the limited amplitude of an auxiliary mass, the following control laws have been developed: gain-scheduling control laws changing the feedback gain according to the magnitude of the disturbance [4], [5], [6], [7], [8]; gain-scheduling control laws with a nonlinear spring [9], [10]; and saturating control laws that limit the input to the actuator [11], [12]. These have the drawback that the control algorithm is complex. Also, except the control law in [4] that uses the technique developed in [13], they do not theoretically assure that the auxiliary mass satises the amplitude constraint or that the closed-loop system is asymptotically stable. This paper proposes a control law for an AMD that effectively suppress the vibrations of a multi-degree-offreedom structure system under the amplitude constraint of the auxiliary mass. The proposed control law has the following features: 1) The control algorithm is simple; the control law is constructed based on a linear saturating control and can be computed as easily as an LQ optimal regulator. Thus, it can also be applied to systems where an LQ optimal regulator is already implemented. 2) The asymptotic stability of the control system is theoretically guaranteed. Moreover, it provides a good control performance. The effectiveness of the control law is examined by simulations.

II. M ATHEMATICAL MODEL OF THE CONTROLLED

OBJECT AND PROBLEM STATEMENT

Fig.1 shows the analytical model of the structural system considered in this study, which consists of rigid oors and linearly elastic side plates and has an AMD on the top oor. The symbol mi represents the mass of the ith oor, ki the spring constant of the ith paired plates, ma the auxiliary mass of the AMD, u the force generated by the actuator of the AMD. The quantities qi and qa are the displacements of the ith oor and the auxiliary mass from the nominal position, respectively. It is assumed that there are no dissipative forces. The equations of motion are obtained from the balance of

978-1-4244-3492-3/09/$25.00 2009 IEEE

912

-u u kn mn

qa ma qn

m2 k2 k1

q2 q1

m1

and the u making (7) hold is obtained from (2), (3), (4), and (7) as { } kn ma mn (qn1 qn ) + r u= mn + ma mn { } kn 2 1 ma mn 1 (qn1 qn ) 2 r r + 2 v . = mn + ma mn T T T (8) Then the following holds for the 1-norm of G(s), denoted G(s)1 : G(s)1 =
0

|g(t)|dt = 1

(9)

Fig. 1.

n-degree-of-freedom structural system with an AMD.

forces as m1 q1 + (k1 + k2 )q1 k2 q2 = 0 m2 q2 k2 q1 + (k2 + k3 )q2 k3 q3 = 0 . . . mn1 qn1 kn qn2 + (kn1 + kn )qn1 kn qn = 0 (1) mn qn kn qn1 + kn qn = u (2) ma qa = u. (3)

where g(t) is the impulse response of G(s). Let R be the set of all solutions of (7), [r(t) r(t)] , t 0, reachable from the origin by some input v satisfying |v(t)| a. Thanks to (9), condition (5) is satised if the following two conditions hold (see Appendix I). [r(0) r(0)] R |v(t)| a, t 0 Also, substitution of (8) into (2) yields (mn + ma )n + kn (qn qn1 ) = ma r. q Now dene q := [ qn ] (12) (10) (11)

q1

Let r denote the relative displacement between the auxiliary mass and the top oor: r = qa qn . Moreover, r is supposed to be constrained as |r(t)| a, t 0 (5) (4)

Then the equation of motion (1) and (12) are written in terms of q as M q + Kq = L r (13) where K= M = diag {m1 , . . . , mn1 , mn + ma } k1 + k2 k2 0 . . . 0 L= [ k2 k2 + k3 .. . 0 0 k3 .. . .. . 0 .. . 0 . . . 0 kn kn

where a > 0 is the maximum allowable amplitude of the auxiliary mass. The problem is to nd a control law for the AMD that quickly suppresses the vibrations of the structural system generated by some impulsive disturbance. III. D ESIGN METHOD A. Reduction to a problem with constrained input Let v(t) be a new input for the AMD, r(s) and v(s) denote the Laplace transforms of r(t) and v(t), respectively, and G(s) denote the transfer function from v(s) to r(s). The input u is given so that G(s) takes the form G(s) = 1 r(s) = v(s) (1 + T s)2 (6)

kn1 + kn kn ] 0 ma .

The mathematical model of the structural system with AMD is represented by (7) and (13), or 2 1 1 r r + 2v 2 T T T ( ) 1 2 1 1 1 q = M Kq + M L 2 r r + 2 v . T T T (14) Dene the state x as [ ] x := r r q q . (15) r = Then (14) is represented by the following state equation: x = Ax + Bv (16)

with T > 0 being a design parameter. The time-domain expression of (6) is 1 2 1 r = 2r r + 2v T T T (7)

913

where A= with

A11 A21 A11 = [

0 A22 [

] , B= 1 2 T

B1 B2

Hence, with the change of coordinates the system has been decomposed into the two subsystems: ws = A11 ws + B1 v wu = A22 wu + B2 v. (22) (23)

] ] [ 01 A21 = 2 M 1 L T 2 T [ ] 0 In A22 = M 1 K 0 ] [ ] [ 0 0 , B2 = . B1 = 1 M 1 L T12 T2 Here In denotes the n n identity matrix. A solution of the problem will be obtained by solving the more tractable problem: nd a control law that asymptotically stabilizes the system (16) under conditions (10) and (11). B. Stabilization by partial state feedback The problem in Section III-A will be solved after being reduced to a much easier one, by decomposing the system (16) into the stable and unstable subsystems by a change of coordinates, where a control law is to be found that asymptotically stabilizes the unstable subsystem under the constraint of the input v. Let S be a coordinate transform matrix such that the change of coordinates w = Sx transforms the state equation (16) into the following w = Aw + Bv where 0 A = SAS 1 = A22 [ ] = SB = B1 . B B2 [ A11 0 ] (18) (17)

0 T12

The ws subsystem is asymptotically stable, while the wu subsystem is unstable (marginally stable). Thus, a control law v = f1 (wu ) that asymptotically stabilizes the wu subsystem under |v(t)| a asymptotically stabilizes the whole system; moreover if condition (10) also holds, then the control law satises the constraint on r, i.e., (5), and thus it is also a solution of the original problem. A saturating control that globally asymptotically stabilizes the wu subsystem under condition |v(t)| a is given by (see Appendix III) v = sat(B2 P wu , a) (24)

where is any positive constant and P is a positive denite matrix satisfying the following equation: A P + P A22 = 0. 22 (25)

Here sat() is the saturating function with amplitude a dened by sat(, a) = sgn() min{||, a}. (26)

The u obtained by substituting v in (24) into (8) is the proposed control law. C. Computing P satisfying (25) The matrix A22 has 2n distinct purely imaginary eigen values 1 , 1 , . . . , n , n (see Appendix III). Let U be a coordinate transform matrix diagonalizing A22 ; that is, U is a matrix formed upon using the associated set of eigenvectors of A22 as columns. Write A22 = U 1 A22 U = diag{1 , 1 , . . . , n , n }. Note that for any i > 0, i = 1, . . . , n, the following P is a positive denite matrix satisfying (25): P = U 1 diag{1 , 1 , . . . , n , n }U 1 .

(27)

Such a matrix S is given by (see Appendix II) [ ] I2 0 S= X I2n where X is the solution of the Sylvester equation A22 X + XA11 + A21 = 0.

In fact, multiplying A P + P A22 on the left by U and on 22 the right by U gives (19) U A U 1 U P U + U P U U 1 A22 U 22 = diag{1 , 1 , . . . , n , n }diag{1 , 1 , . . . , n , n } (20) +diag{1 , 1 , . . . , n , n }diag{1 , 1 , . . . , n , n } = 0. Various simulations show that a controller providing a good control performance can often be obtained by choosing i , i = 1, . . . , n so that B2 P approximates an LQ optimal regulator gain used in the design of a low-and-high-gain saturating regulator [15]. First consider the Riccati equation A P + P A22 P B2 B2 P + Q = 0 22 (28)

Since A11 and A22 have no common eigenvalues, (20) can be solved uniquely [14]. The state w is partitioned as [ ] ws w= (21) wu where ws = [ ] [ ]

w1

w2

, wu =

w3

w2n+2

914

where Q is a positive denite matrix and is a sufciently small positive number. Let B2 , P , and Q denote the ma2 , P , and Q, respectively, in the trices corresponding to B coordinate system where A22 is diagonalized. That is, B2 = U 1 B2 , P = U P U, Q = U QU. In this coordinate system (28) is written as 22 A P + P A22 P B2 B2 P + Q = 0. (29)

IV. N UMERICAL RESULTS Consider a 4-story building with an AMD on the top oor. The parameters of the structural system are given as n=4 mi = 2000 kg, ki = 250 kN/m, i = 1, . . . , 4 ma = 50 kg, a = 0.5 m. Also, the initial state variables are given by r(0) = 0, r(0) = 0 qi (0) = 0, i = 1, 2, 3, q4 (0) = 0.2 m qi (0) = 0, i = 1, . . . , 4. Figs.2 through 4 show the numerical results for T = 0.02, 0.1, and 0.15 s, respectively. In each gure the relative displacement between the auxiliary mass and the top oor, r, and the displacement of the top oor, q4 , are shown. With T being as small as 0.02 s the auxiliary mass is moved almost from end to end in the allowable range to suppress the vibrations of the structure effectively. As T becomes larger, the control performance becomes lower. This is because as T is made to be larger, the speed of response of the ws subsystem becomes slower and moreover the wu subsystem becomes more difcult to control. The degree of controllability of the wu subsystem varies depending on T . Note from (17)[ that if r(0) = 0 and r(0) = 0, then ] w(0) = Sx(0) = q(0) q(0) , which is not affected by T . Thus, the degree of controllability may be determined by the transformed state equation (23), or by A22 and B2 , but only B2 depends on T . So the magnitude of B2 may be a measure of the degree of controllability. Fig.5 shows that B2 rapidly decreases as T increases, which illustrates the above consideration.
0.8 0.4

Since (A22 , B2 ) is controllable and Q is a positive denite matrix, (28) has a unique positive denite solution P , for which the following relation holds [15].
0

lim P = 0

Therefore, from (29), if Q is chosen as a diagonal matrix Q = diag{1 , 1 , . . . , n , n }, i > 0 then for a sufciently small , the positive denite solution P becomes a diagonally dominant matrix. Denote the diagonal elements of P by 1 , 1 , . . . , n , n , and construct P as follows: P = U 1 diag{1 , 1 , . . . , n , n }U 1 .

(30)

Since i > 0, i = 1, . . . , n, P is a positive denite matrix and, as stated above, it satises (25); also, the matrix P approximates P . The matrix Q is given, based on the idea that a larger weighting coefcient is given for a slower mode, which corresponds to a smaller |i |, as follows: { } 1 1 1 1 = diag Q , , ..., , . (31) |1 |2 |1 |2 |n |2 |n |2 The parameter is designed so that the speed of response in the linear range is maximized. That is, it is designed in such a way that the maximum real part of the set of eigenvalues of the matrix A22 B2 B2 P is minimized. The above procedure for designing P and is summarized as follows: Algorithm for computing P and 1) Compute the eigenvalues and eigenvectors of A22 , and form the coordinate transform matrix U using the set of eigenvectors as columns. 2) Obtain Q using (31). 3) Compute Q by Q = U 1 QU 1 . 4) Obtain the positive denite solution P of (28) for a sufciently small > 0. 5) Compute P by P = U P U . 6) Obtain P using (30). 7) Find a > 0 minimizing the maximum real part of the set of eigenvalues of the matrix A22 B2 B2 P by one-dimensional search.

r [m]

0 -0.4 -0.8 0 0.2 0.1 5 10 15 20 25

t [s]

q4 [m]

0 -0.1 -0.2

10

15

20

25

t [s]

Fig. 2.

Numerical results for T = 0.02s.

915

0.8 0.4

0 -0.4 -0.8 0 0.2 0.1 5 10 15 20 25

t [s]

0 -0.1 -0.2

10

15

20

25

t [s]

Fig. 3.
0.8 0.4

Numerical results for T = 0.1s.

r [m]

saturating one and is similar to the saturating control obtained by the low-and-high-gain approach, proposed by Saberi et al. While Saberi et al.s control law can be applied to a bounded set in the state space, the proposed one assures the global asymptotic stability. If the time-optimal control under |v(t)| a for wu subsystem, which can explicitly be obtained when n = 1 [16], is used as the control v and moreover T is made to approach 0, then the structure system is controlled timeoptimally under the condition |r(t)| a. However, such a control has no robustness against observation noise and modeling errors, and requires an innitely large input u when T approaches 0. Therefore, a practical control law was constructed by using a saturating control to stabilize the wu subsystem and choosing T that is not too small. For persistent disturbances, the authors recommend that the saturating control is used mainly in the linear range (by adjusting ). Then the control law approximates an LQ optimal control law, which has well been tested in practical applications. A PPENDIX I P ROOF OF THE FACT THAT IF (10) AND (11) HOLD , THEN (5) IS SATISFIED . Conditions (10) and (11) can be replaced by [ ] r() r() = 0, |v(t)| a, t > . From these, r(t) is computed as t r(t) = g(t )v( )d.

q4 [m]

r [m]

0 -0.4 -0.8 0 0.2 0.1 5 10 15 20 25

t [s]

q4 [m]

0 -0.1 -0.2

10

15

20

25

t [s]

Fig. 4.
4

Numerical results for T = 0.15s.

Use of the change of variable = t yields r(t) = g()v(t )d.

0

From this and (9), the following inequality is obtained. |r(t)| |g()| |v(t )|d a
0

A PPENDIX II P ROOF OF THE FACT THAT S IS GIVEN BY (19) Let S has the form [ ] I2 0 S= X I2n where X is unknown. From the inverse matrix formula, we have [ ] I2 0 1 S = . X I2n Then SAS 1 is computed as [ ][ ][ ] I2 0 A11 0 I2 0 1 SAS = X I2n A21 A22 X I2n [ ] A11 0 = . A22 X + XA11 + A21 A22 Therefore, if X satises (20), then the following holds. [ ] A11 0 1 SAS = 0 A22

~ || B 2 ||

0 0 0.2 0.4 T [s] 0.6 0.8 1

Fig. 5.

T versus B2 .

V. C ONCLUDING REMARKS For a multi-degree-of-freedom structural system with an AMD, a control law of the AMD has been proposed that realizes an effective vibration suppression control under the AMDs stroke constraint. The control law has the form of a

916

A PPENDIX III P ROOF OF THE FACT THAT THE CONTROL LAW (24) GLOBALLY ASYMPTOTICALLY STABILIZES THE wu
SUBSYSTEM

= rank

B2

A22 B2

A2n1 B2 22

From the structures of A22 and B2 it can be seen that 2 ) is controllable. Also, since the autonomous system (A22 , B wu = A22 wu is an undamped oscillatory system, its solu tions do not diverge. That is, let i denote the eigenvalue of A; modes such as tk ei t do not exist, which means that when A22 is transformed into a Jordan canonical form, there are no Jordan blocks of order greater than 1. Moreover, since (A22 , B2 ) is controllable with a single input, A22 does not have any multiple eigenvalues; so A22 has 2n distinct eigenvalues and is diagonalizable. This means that (25) has a positive denite solution; see (27). Dene the following positive denite function using a positive denite solution P of (25):
V := wu P wu .

That is, (A22 , B2 ) is uncontrollable, which contradicts the fact. Therefore, V 0 holds only if wu 0. From LaSalles invariance principle [17], the system is asymptotically stable. R EFERENCES
[1] B. F. Spencer Jr. and M. K. Sain: Controlling buildings: A new frontier in feedback, Control Systems Magazine, IEEE, Vol.17, Issue 6, pp. 1935, Dec. 1997. [2] Kazuto Seto: Vibration Control of Structures, Corona Publishing Co., Ltd., 2006(in Japanese). [3] Architectural Institute of Japan Ed.: Active and Semiactive Control for Buildings: State of the Art, Maruzen Co., Ltd., 2006(in Japanese). [4] A. Nishitani, Y. Nitta and N. Yamada, Variable gain-based structural control considering the limit of AMD movement, Proc. of the 35th IEEE Conf. on Decision and Control, pp.185190, Dec. 1996. [5] I. Nagashima and Y. Shinozaki, Variable gain feedback control technique of active mass damper and its application to hybrid structural control, Earthquake Engineering and Structural Dynamics, Vol.26, pp. 815838, Aug. 1997. [6] M. Yamamoto and Y. Suzuki, Experimental study on seismic response control of full-scale structure based on pole assignment algorithm considering AMD with stroke limitation, J. Struct. Constr. Eng., Architectural Institute of Japan, No. 514, pp. 127132, Dec. 1998(in Japanese). [7] S. Fukuzumi, A. Sone and D. Iba, Vibration control of structure considering stroke constraint of auxiliary mass of TMD, ASME, Pressure Vessels and Piping Division, PVP-Vol. 428-2, pp. 163-169, Jul. 2001. [8] J. Chunyan, L. Huajun and M. Qingmin, Active control strategy for offshore structures accounting for AMD constraints, High Technology Letters, Vol. 10, No. 4, pp. 6368, Dec. 2004. [9] H. Nishimura, N. Oie and K. Takagi, Vibration control of a structure by using of an active dynamic vibration absorber taking account of actuator constraints, Trans. of the Japan Society of Mechanical Engineers, Ser. C, Vol. 66, No. 641, pp. 5359, Jan. 2000(in Japanese). [10] N. Oie, H. Nishimura and S. Shimodaira, Vibration control of a structure by using of an active dynamic vibration absorber taking account of actuator constraints: Experimental verication for a multidegree-of-freedom structure, Trans. of the Japan Society of Mechanical Engineers, Ser. C, Vol. 68, No. 665, pp. 5259, Jan. 2002(in Japanese). [11] B. Bhartia and I. Nagashima, Saturation control of buildings with AMD of bounded capacities, Collection of Technical Papers AIAA/ASME Structures, Structural Dynamics and Materials Conference (pt 6), pp. 3423-3429, Apr. 1993. [12] J. Mongkol, B. Bhartia and Y. Fujino, On linear-saturation (LS) control of buildings, Earthquake Engineering and Structural Dynamics, Vol. 25, pp. 13531371, Dec. 1996. [13] K.Yoshida, Y.Nishimura and Y.Yonezawa, Variable gain feedback control for linear sampled-data systems with bounded control, Control - Theory and Advanced Technology, Vol. 2, No. 2, pp. 313-323, Jun. 1986. [14] F.R. Gantmacher, The theory of matrices Vol.I, Chelsea Publishing Company, p.225, 1959. [15] A. Saberi, A. Stoorvogel and P. Sannuti, Control of linear systems with regulation and input constraints, Springer-Verlag London Limited, pp. 75108, 2000. [16] E.P. Ryan, Optimal relay and saturating control system synthesis, IEE Control Engineering Series 14, Peter Peregrinus Ltd., p.118, 1982. [17] H.K. Khalil, Nonlinear Systems 3rd Ed., Upper Saddle River, NJ: Prentice Hall, pp.126128, 2002.

The rate of change of V is

V = wu (A P + P A22 )wu + 2wu P B2 v 22 = 2wu P B2 sat(B2 P wu , a) 0

where (25) was used. From the above equation we see that V 0 holds only if B2 P wu 0. Now suppose that wu (0) = 0 and B2 P wu (t) 0, t 0. Then we have v 0, and the differential equation for wu is obtained from (23) as wu = A22 wu . From the assumption we have the following (by obtaining the kth time derivative of B2 P w(t) and putting t = 0, where k = 0, . . . , 2n 1): B2 P wu (0) = 0 B2 P A22 wu (0) = 0 . . . B2 P A2n1 wu (0) = 0. 22 Also, from (25) we get P A22 = A P ; substitution of this 22 into the above equations gives B2 P wu (0) = 0 B2 A P wu (0) = 0 22 . . .
2n1 B2 A P wu (0) = 0. 22

Since P is nonsingular, the above equations mean that B2 B2 A 22 2n > rank . . . 2n1 B A
2 22

917