515

Control of vibration in civil structures

K Seto
Department of Mechanical Engineering, College of Science and Technology, 1-8 Kanda Surugadai, Chiyoda-ku,
Tokyo, Japan. E-mail: [email protected]

Abstract: This paper reports recent trends in active vibration control mainly as developed in Japan
for civil structures. Firstly, it classifies vibration control methods and controllers, especially active
dynamic absorbers that are widely used in mechanical and civil engineering. Secondly, it addresses
basic problems in the control of vibration of flexible structures such as formulating the reduced-
order model required for designing vibration controllers, the correct arranging of sensors and
actuators, and how to prevent spillover instability. Finally, the practical use of control theories such
as linear–quadratic control theory, H control theory, neural network theory and other topics are
2
discussed.

Keywords: active vibration control, vibration controller, control theory, modelling, sensor and actuator

1 INTRODUCTION method for controlling vibration. Although lighter-weight

bodies are needed to decrease consumption in auto-
Vibration control prevents resonance and unstable mobiles, it becomes necessary to control the vibration of
vibration, as well as instigating quick suppression of these somewhat elastic automobile bodies. Increasingly
transient vibration by complementing the shortage of lighter and taller buildings have been constructed to meet
internal damping from the outside or by generating demands for lower costs. Such buildings involve the issue
internally a force that cancels external forces. As a con- of swaying caused by strong winds. Behind this recent
ventional vibration control method, passive-type vibration activity of research could be the fact that the modern
control has been the main focus and does not complement control theory has before been considered difficult to
any energy feeding from the outside. The first active understand and thus not likely to be put to practical
vibration control device introduced in Japan was com- application, but recently it has become applicable to con-
posed of a sensor and a controller in the form of a servo trol system design because of ongoing research, advances
damper. The control method was proposed in 1970 by in computer application techniques and the appearance
Tominari et al. to prevent the chattering vibration of of effective software packages such as MATLAB.
machine tools [1]. The servo damper was followed by a The dynamic behaviours of systems to be vibration
torsion servo damper devised by Okada and Nakada [2]. controlled have come to be analysed in detail, experi-
Tanaka et al. sought to establish a control system design mentally as well as theoretically [6 ] and the introduction
method for a servo damper incorporating modern control of a fast Fourier transform (FFT ) analyser has enabled
theory [3]. While this method proved to have excellent easier examination of vibration control results. These
vibration-controlling performance, it was not adopted in remarkable advances in hardware and software can be
practical applications because it was not easily understood itemized as follows:
and lacked reliability.
(a) development of software for control system design;
Active vibration control has recently become a key
(b) provision of high-speed computers and the intro-
technology for maintaining and enhancing the capability,
duction of digital signal processes (DSPs), etc.;
performance and dwelling properties across a wide range
(c) development of new active vibration control devices
of engineering fields [4, 5]. For example, with industrial
and actuators;
robots, lighter weight, energy savings, higher speeds and
(d) advance in the computer aided vibration analysing
greater accuracy are increasingly demanded. An obstacle
methods and experimental modal analysers;
to achieving such properties is the availability of light
(e) appearance of a new control system design method.
and flexible arms, creating the need to devise a practical
In this paper, control devices whose design methods
The MS was received on 8 September 2003 and was accepted after have been developed are presented as active devices.
revision for publication on 4 August 2004. Current trends and the complex problems of active
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
516 K SETO

vibration control are described. A procedure for making into passive, semiactive, active and hybrid types. In order
a reduced-order model for flexible structure is presented. for the actuator to generate a vibration control force, it
Finally, the results from applications of newly released is necessary for it in some way to receive a reaction force,
and recently trialed control system design methods since action is followed by reaction. Vibration control
including improved versions of the linear–quadratic (LQ) methods are also classified according to the way that the
control theory, H control theory, sliding-mode con- reaction force is received [7]. The classes are as follows:
2
trol and neural networks, are presented. Further, recent (a) method 1 using the reaction of a fixed point;
developments in simultaneous control of motion and (b) method 2 using the reaction of an auxiliary mass;
vibration will be introduced. (c) method 3 using the reaction of an auxiliary structure.
Figure 1 indicates a schematic view of the vibration
2 ACTIVE VIBRATION CONTROLLER control methods. The controlled objects are simplified
into systems with one degree of freedom (DOF ).
2.1 Classification of vibration controllers Method 1 allows for the simplest construction of a
control system of the three methods, provided that there
The main purpose of vibration control is to stabilize a is a fixed point near the actuator’s mounting point. For
controlled object, where a high reliability is required in instance, active vibration isolation systems and active
a vibration control device. Passive vibration control suspension systems for cars fall into this category. If
devices have traditionally been used, because they do there is a fixed point to mount the actuator, this method
not require an energy feed and are therefore free from is applicable. Precision vibration control techniques to
the risk of generating unstable states. However, passive isolate environmental vibrations based on this method
vibration control devices have no sensors and cannot will be applicable in many fields, particularly in pro-
respond to variation in parameters of the controlled cessing facilities for extra-precision measurement using
object or controlling device. This has resulted in the laser devices [8], active suspensions of vehicles [9], micro-
evolution of a new type of vibration control device called gravity environments, etc. However, if a fixed point is
an active vibration controller, which is equipped with not readily available for mounting the actuator, then one
sensors, an actuator and a controller that requires power. of the other two methods will be needed to receive the
Therefore, the vibration controllers are roughly classified reaction force.

Fig. 1 Classification of vibration methods

Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 517

Method 2 uses the reaction inertia force of an auxiliary is architecturally preferred, it is structurally weak because
mass to produce a vibration control force in an actuator. of its susceptibility to strong winds. By placing an active
Several types of dynamic absorber that use this method vibration control device in the upper portion of the two
are in wide use in the field of engineering, since it has buildings as shown in Fig. 2b, vibrations generated by
the advantage of the actuator being able be mounted at wind excitation, the Kallman vortex, are eliminated.
any point. Method 2 is classified, as shown in Fig. 1, Method 3 is also classified, as shown in Fig. 1, into
into passive, semiactive, active and hybrid types. In the passive, semiactive, active and hybrid types. The active
field of mechanical engineering, the passive type is and hybrid types are proposed by the present author,
generally called a dynamic absorber and can achieve an and this method is called the connected control method
excellent controlling effect by keeping three elements, (CCM ). An application of this method is shown later.
namely the mass, the spring and the damper, in optimum
adjustment [10, 11]. In the field of civil engineering, this
is called a tuned mass damper. However, since the 2.2 Construction of active dynamic absorber
dynamic absorber does not have its own sensors, it
The advantages of the active dynamic absorber are that
cannot adapt to changes in the characteristics, if any, in
its vibration control effect is not only excellent and
the control object or in itself, and it often has difficulty
robust, but it is able to on its own control the vibration
effectively controlling the vibration. This problem is
of many-DOF systems [13].
particularly serious where the mass ratio is small. There
The active dynamic absorber is formed by substituting
are two ways to solve this problem. One way is to
the damper of the dynamic absorber with an actuator that
change the characteristics of the spring or the damper
is controlled by a sensor and a controller. Because the
of the passive dynamic absorber, and to ensure that
auxiliary mass is driven by an active device, it is called
the optimum adjustment condition is maintained. This
an active mass. Hydraulic-type active dynamic absorbers
method is called a semiactive dynamic absorber, although
are applicable to vibration control of a control object
fundamentally it belongs to the passive-type category.
that requires a powerful but compact vibration controller.
The other way is to use an active dynamic absorber with
To eliminate self-exciting chatter vibration occurring in
controller, sensors and actuators. In the field of civil
machine tools, an active dynamic absorber called a servo
engineering, active and hybrid types are called an active
damper shown in Fig. 3 has been proposed [1].
mass damper (AMD) and a hybrid mass damper (HMD)
A typical active dynamic absorber linearly driven by
respectively.
an electromagnetic force and its control system are shown
Method 3 generates a control force with an actuator
in Fig. 4. Based on the driving method of the auxiliary
mounted on a structure located parallel to the flexible
structure, to control the vibration of that structure.
An application of this system is shown in Fig. 2 [12].
The control device can be easily constructed, because it
only requires insertion of an actuator between the two
structures. Figure 2a illustrates adjacent but independently
constructed multi-storey buildings. Although this design

Fig. 3 Schematic view of a servo damper

Fig. 2 Plan of the method that uses the reaction of an auxiliary Fig. 4 Active dynamic absorber driven by an electromagnetic
structure force

I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
518 K SETO

Fig. 5 Landmark Tower equipped with two HMDs ( left) and the outside view (right)

mass, linear drive and motor drive systems are avail-

able. The former method directly moves the auxiliary
mass, and the latter indirectly moves the auxiliary mass
by rotating the ball screw with a servo motor [14].
From the method of supporting the auxiliary mass, the
spring, guide and pendulum supports can be ascertained.
Frequently, an active mass is supported without friction.
Typical examples include a guide roller, pendulum and
spring support.
Recently, a number of high-rise buildings (over 60)
and huge bridge towers equipped with AMDs or HMDs
have been constructed in Japan [15]. Although active
vibration control techniques have been actively sought
as countermeasures against strong winds, some problems Fig. 6 Multi-step pendulum-type HMD
remain unresolved in respect of their reliability and
energy savings. A practical solution to this problem is
the hybrid vibration control method, which combines Tower. A hanging pendulum is driven by a motor
both passive and active vibration control mechanisms operating linearly and independently in two horizontal
into a single control unit. For example, the Yokohama directions. It is possible to tune the natural frequency of
Landmark Tower (the building 300 m high shown in the pendulum to that of the building. An important
Fig. 5) was completed in 1993 adopting this hybrid feature of the arrangement is the compact multi-step
method. An HMD arranges an actuator and a damper pendulum to tune the natural frequencies of the tower.
in parallel. Because vibration energy is absorbed by the The counter-electromotive force produced by the rotation
damper, to a certain degree some reliability is ensured, of the motor serves as a damper.
even though a failure could be generated by an active Figure 7 illustrates a system with an active mass,
system. Two recent applications are presented by way of which is formed by cutting the arc of a circle and driven
example. by a servo motor [17]. Because the motions of an arc of
a circle have a constant period, the natural frequency
of the mass can be set to any value through proper
2.3 Examples of application in structural control selection of the radius L. The application of damping in
the direction of motion forms a hybrid vibration control
2.3.1 Wind excitation control for buildings using HMD
device that consists of motor drives and damping forces.
Figure 6 shows the outline of the vibration control The counter-electromotive force of the motor serves as
device [16 ] incorporated in the Yokohama Landmark a damper. In addition, passive operation is provided in
Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 519

2.3.3 Vibration control of triple towers using the

connected control method
An outline of Harumi Triton Square is sketched in
Fig. 10. The triple towers are denoted X, Y and Z in
order of height and are arranged at 45° to tower Y.
Towers X, Y and Z are 200, 175 and 145 m in height,
with 45, 39 and 30 storeys respectively. At the design
stage, it became clear that a complex vortex excitation
from winds was of concern at the first bending mode of
the triple towers [19].
To control the first bending mode of each tower, it
Fig. 7 Roller guide-type HMD was necessary to use two AMDs, for the two directions;
therefore, in total, six AMDs would be required if a
conventional system was employed. Using our proposed
the event of power failure. Figure 8 shows an improved CCM, the minimum requirements are two controlled
HMD called a V-shaped HMD, because the radius L is bridges (devices). The company made a decision to use
selected by changing the angle of the V-shaped rail. the CCM and the two controlled bridges as shown in
Tokyo Park Tower is equipped with three of this type Fig. 11. Tower X is connected with tower Y by one of the
of HMD [18]. controlled bridges at the 39th storey. Similarly tower Z
is connected with tower Y at the 30th storey. Each tower
independently gains control force in two directions,
2.3.2 Active vibration control of large bridge towers x and y.
under construction The actual active controlled bridges were designed and
Active vibration control of large bridge towers while developed by Ishikawajima Heavy Industry [20]. The
under construction is a common practice in Japan [15] construction is schematically shown in Fig. 11. An active
to reduce the risk of destruction by strong winds during controlled bridge consists of an outer tube, an inner tube,
the course of construction and is shown in Fig. 9a. For a slide guide between each tube, a two-bowl-screws
the purpose, AMDs and HMDs are commonly applied. driving mechanism and two a.c. servomotors. The a.c.
Once the bridge towers are connected by a suspension servomotors, mounted one to the upper side and one to
rope, the vibration controllers have finished their job the lower of the inner tube, drive the bowl screw to move
and are removed, as shown in Fig. 9b. the outer tube in a forward and back direction. Therefore
As a tendency to construct more reliable but lighter both ends of the tube are expanded and contracted by
towers increases, it becomes insufficient to consider just the active control of the a.c. servomotors. Since both
the first mode, since vibrations of higher modes also raise ends of the tube are connected between the two towers
problems regarding structural strength. Generally these through bowl joints, these towers are moved smoothly.
flexible towers are distributed parameter systems and, to An outside view of the active controlled bridge con-
design a feedback controller for such a system using necting towers Y and Z is shown in Fig. 11. The first
linear control theory, it is necessary to obtain a suitable triple-towers complex connected with two active con-
reduced-order model by neglecting high-frequency modes. trolled bridges using our proposed CCM was realized in
downtown Tokyo in 2001, as shown in Fig. 12.

3 IMPORTANT ITEMS FOR ACTIVE VIBRATION

CONTROL

Interesting examples of recent applications of active

vibration control include:

(a) vibration control of flexible robot manipulators [21];

(b) active vibration control of elastic automobile bodies
[22];
(c) swing control of high-rise buildings and bridge towers
[23, 24];
(d) control of flexible rotors with magnetic bearings
Fig. 8 Outside view of a V-shaped HMD used in Tokyo Park [25, 26 ];
Tower, which has three of this type of HMD (e) vibration control of large space structures [27].
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
520 K SETO

(a) Towers standing independently (b) Towers connected by a suspension rope

Fig. 9 Suspension bridge under construction

Fig. 10 The outline of Harumi Triton Square

Fig. 11 Schematic view of an active controlled bridge and an outside view of the active controlled bridge
connecting towers Y and Z

Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 521

Fig. 12 Outside view of the triple tower connected by active bridges

Vibration control of flexible structures have common controlled and the uncontrolled modes; spillover thus
problems. When active vibration control of flexible caused is called observation spillover.
structure is sought, three items must be individually In Fig. 14, an example of spillover is shown by
considered: an impulse response and a frequency response experi-
mentally obtained on a flexible structure, where the first
(a) problem of creating reduced-order models;
two modes are to be controlled and those above the third
(b) placement of sensors and actuators;
modes (including this mode) are not controlled. While
(c) countermeasures for spillover.
vibration from the first two modes decays quickly under
the control, the vibration of the fourth mode grows with
time, leading to spillover. Naturally, spillover is a very
3.1 Spillover
dangerous phenomenon that can cause local damage
A problem that goes with reduced-order modelling of a or even destroy whole structures. Therefore, avoiding or
structure in active control is the spillover problem. It is suppressing spillover is a very important task to achieve
the phenomenon of interference between the controlled successful control.
modes and the omitted, uncontrolled, higher modes. There are various ways to prevent spillover from
Figure 13 illustrates this relationship. In Fig. 13, the sub- occurring:
script C refers to the controlled modes and the subscript
R to the uncontrolled residual modes. While injecting (a) modal filtering method;
control energy to the controlled modes, an actuator may (b) direct feedback method;
also excite higher modes that are not modelled for con- (c) low-pass filter method;
trol. This is called control spillover. On the other hand, (d) robust control method.
sensors may pick up information related to both the
These main methods can be summarized as follows.
The first method was initially proposed by Balas [28]
and is based on separating the information of the con-
trolled modes from that of the uncontrolled modes. This
method requires the same number of sensors and actuators
as that of the controlled modes. The second method was
also proposed by Balas [29] and requires placing sensors
and actuators at the same locations (collocation). While
the stability of the control system is guaranteed, great
control effects cannot be expected. The third method
uses low-pass filter to cut off either the pick-up signals
or the actuation forces in order to suppress the effects
of the higher uncontrolled modes. It is simple but often
fails to control multiple modes where modal frequencies
Fig. 13 Control and observation spillovers are not sparsely spaced. An improvement on this method
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
522 K SETO

Fig. 14 Example of spillover

is introduced in section 3.2 that follows. The fourth multiple modes of flexible structures. For this purpose,
method is fairly new and has been rapidly developed in a multi-DOF lumped-mass model is derived. Based on
recent years; this method is further discussed later in this reduced-order physical model, a feedback controller
this paper. is designed utilizing filtered LQ control and the H -based
2
control law. In this section, such a method previously
described by the present author is introduced.
3.2 The lumped modelling method The first step of this method involves the deter-
mination of the number and positions of the modelling
3.2.1 An idea crucial for deriving a reduced-order
points. This profoundly relates to the controllability
physical model [30, 31]
and observability problem. It is well known that a
As shown in the previous section, for a lumped-parameter controller design in a state space is only meaningful
system, designing a controller in a physical state space when the system is observable and controllable. For
is straightforward. Unfortunately, most structures are distributed systems, it is not easy to judge this by any
distributed in nature and cannot be regarded as lumped well-established algebraic method. The present author
parameter systems. The modelling method presented has thus proposed a simple method to achieve this by
in this paper, called the reduced-order physical model examining the mode shapes of the structure [6 ]. This can
method (Seto’s method), makes a lumped-parameter be explained by the cantilever beam shown in Fig. 15.
model of a distributed structure and designs a controller Figure 15 shows the first four modes of the beam.
in a physical state space where the states are directly In this figure, w and w are the ith mode shape com-
pi ri
measurable. The method differs from the usually employed ponents at the position of control force and sensor
modal-space control method [32, 33] in which modal state respectively. If they are not zero, the system is con-
estimation is necessary, or the so-called direct output trollable and observable; if they are zero, then it is
feedback control in which sensors are collocated with uncontrollable and unobservable. Obviously, at the nodes
the actuators and a given actuator force is a function of a mode, these components are zero. In other words,
of the sensor output at the same point. The proposed if the control force is applied at a node of a mode, then
method and the skills required for it are not difficult the mode is uncontrollable; if the sensor is located at a
in practice, as verified by experiments in controlling node, the mode is then unobservable. In Fig. 15, the first

Fig. 15 Relationships between controllability/observability and locations of sensors and actuators

Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 523

and second modes are controllable and observable, the system can be obtained, the physical parameters of
third mode is observable but uncontrollable and the the system could then be determined. Unfortunately, the
fourth mode is controllable but unobservable. directly obtained modal matrix from the finite element
Therefore, if the first two modes must be controlled, method (FEM ) modal analysis data is not guaranteed
two modelling points are needed and they can be placed to satisfy equation (3), where both sides should be
at the nodes of the third mode. In this way, the spillover diagonal matrix. Since the mode vectors in W are
problem with this mode can be avoided in the modelling sampled from a larger modal matrix of an FEM model
phase. The present author termed this ‘the structure of the structure or vibration mode shapes obtained by
filter’ [34]. experimental modal analysis, they are not necessarily
The key points of the modeling procedure are as follows: going to satisfy equation (3), which is based on a much
lower order, a lumped parameter model. Substituting W
1. Analyse the vibration mode shapes of the flexible
into equation (3) results in a matrix WWT, whose off-
structures to be controlled.
diagonal positions usually contain non-zero, although
2. Select the nodes of the lowest order non-controlled
small, values. A sensitivity analysis procedure is pro-
vibration mode, place the masses of the lumped para-
posed here to modify the initial matrix W in order to
meter system at the strategically selected nodes and
make WWT diagonal.
make an r-DOF system.
Suppose that there are p different non-zero components
3. Determine the values of masses and spring constants
in the off-diagonal positions of the matrix WWT; an error
using sensitivity analyses.
vector e is defined consisting of these p components,
The next section shows how to construct a lumped namely
model with strategically placed masses at specified points
e={e e e , e )T (5)
on a distributed structure. 1 2 3 p
in which each error e is a function of some modal
i
3.2.2 Modification of a normalized modal matrix shape components of the matrix WWT. The purpose is
In a physical coordinate system, the equation of motion to modify these modal shape components w in order
j
of an r-DOF system with negligible damping can be that the errors e approach zero.
i
expressed as Define the modification vector of W as dW, which is
assumed to change the error vector from its initial value
MẌ+KX=F (1) e to 0; then the correcting procedure can be started from
where M and K are r×r symmetric mass and stiffness the equation
matrices, and X and F are r×1 displacement and force
C D
qe
vectors respectively. From the solution of the eigenvalue dW=0−e=−e (6)
qW
problem given by this equation of motion, modal matrix
W is obtained. The relationship between the mass and The modification vector can be obtained directly from
stiffness matrices of the physical and modal domains are equation (6) using the pseudo-inverse matrix as follows:
given by WTMW=m and WTKW=k.

A B CA BA B D
qe T qe qe T −1
It is reasonable to assign the normalized mode shape dW= (−e) (7)
values of the real structure at the locations of the model- qW qW qW
ling points, which can be obtained through numerical or and the new vector is calculated by
experimental modal analyses, to that of the lumped
system: W+dW
W (8)

w (r ) Usually iterative calculation is necessary in this procedure

w (i )= n i (2) so that e converges to 0. After certain iteration when a
n √m
preset accuracy has been attained, a final modified modal
where r refers to the position of the ith modelling point matrix can be obtained which will satisfy the diagonal
i
on the structure. conditions of equation (8). Consequently, M and K can
Normalizing the modal matrix W yields the following be obtained.
set of relationships for the mass matrix M and stiffness
matrix K of the physical domain: 3.3.3 Application to a bridge tower under construction
M=(WWT)−1 (3) Figure 16 shows a model bridge tower connected with a
model crane tower. The crane tower is connected to the
K=(WT)−1V2V−1 (4)
bridge tower during construction. Modal analysis of
where V2 is a diagonal matrix whose non-zero diagonal the structure is first carried out by an FEM. Figure 17
elements are given by the square of the natural frequencies illustrates the obtained first eight vibration modes of
of each mode. Equations (3) and (4) suggest that, if the model structure, which were found to be the same
the normalized modal matrix of a lumped parameter mode shapes as those of the real structure. The first five
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
524 K SETO

are candidates for the modelling points. The normalized

modal vector components of the structure calculated by
FEM at the modelling points are presumably defined as
the mode vector components of the lumped five-DOF
model, as denoted by the symbols in Fig. 18, namely

C D
w w w w w
11 12 13 14 15
w w w w w
21 22 23 24 25
W= w w w w w (9)
31 32 33 34 35
w w w w w
41 42 43 44 45
w w w w w
51 52 53 54 55
3.3.4 Five-DOF modelling of a scaled bridge tower
structure with a crane tower
The above procedure, which determines the physical
Fig. 16 Model structure of a flexible bridge tower parameters of the lumped parameter system from the
available modal data of distributed parameter system, is
called an inverse modal analysis approach. Figure 18
vividly shows the obtained five-DOF lumped-parameter
model. Two hybrid dynamic absorbers are also shown
connected to the mass points 3 and 4 respectively. They
are the actuators employed in this paper. This discrete
model will make it easy to design a controller in a
physical state space where the states are directly
measurable.

4 DESIGN OF CONTROL SYSTEM

4.1 Theory for active vibration control

There are several control theories which need mention-
ing: classical control theory was initiated in the 1950s;
modem control theory came about early in the 1960s
and has reached practical applications; new wave robust

Fig. 17 Modal shapes of the structure

modes, including bending and twisting modes, are those

needing attention. By choosing five modelling points on
the structure, and assuming appropriate lumped masses
and stiffnesses at and between these points, a discrete
model can be created. Evidently, the modelling points
should be able to reflect the motion of these five modes
as well as possible. In this case, the five modelling points
are chosen to be located at the nodes of the sixth mode,
as shown in Fig. 17, with a plus that the sixth mode will
exert no influence on the model [7]. It should be noted
that choosing the modelling points is not based on the
continuous structure coordinate, but on the FEM model
of the structure; i.e. only the nodes of the FEM model
(here the nodes do not mean the places of zero vibration) Fig. 18 The lumped-parameter model

Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 525

control theory focuses on H control theory and learning Then the system can be described by the state equation
2
control centring in a neural network.
Ẋ =A X +B U
Classical control theory represented by proportional– c c c c c
integral–derivative control continues to provide great y =C X
c c c
utility in practical applications for conventional control (12)
systems but does not suit vibration control in many-DOF
systems such as flexible structures, since it is difficult to These system matrices A and B are easily obtained from
c c
deal with multi-variable systems. the lumped-mass model shown in Fig. 18.
Modem control theory which is based upon optimum As stated before, by placing sensors and actuators
control theory was at first very difficult to apply for users at the nodes of certain modes, the control system
because it was mathematically unintelligible to most. will produce no influence on these modes due to the
Assisted by progress in computer utilization techniques, unobservability and/or uncontrollability characteristics.
this control method has become widely used in the field It seems these modes are cut off by a filter. This is
of vibration control. Optimum control theory deals regarded as a ‘structural filter’. Usually, it is possible to
with linear state-variable feedback control system. It avoid the spillover problem of one or two modes by this
is emphasized that state-variable feedback is the most ‘structural filter’, which creates a favourable frequency
effective way to maintain a desired process balance by allowance for the choice of the cut-off frequency of
achieving desired closed-loop system eigenvalues. In this a low-pass filter. The above consideration is clearly
theory, the state feedback gain is determined to minimize illustrated in Fig. 19. In this figure, a low-pass filter is
the performance index which is a weighted sum of used to prevent spillover caused by neglected higher
squared errors and controlling inputs. Because the per- modes. Thus, designing an optimal controller using
formance index is expressed in an LQ form, it is called LQ theory becomes a fairly standard problem. Using LQ
LQ control theory. To obtain a stabilizing system, robust theory, the optimal feedback gain K=[ K , K ] is easily
f c
control theories are proposed. designed by an extended system combining the control
object denoted by subscript c with a low-pass filter
denoted by subscript f. It should be noted that, owing
4.2 Control system designing using LQ control to practical constraints, the states at mass point 5 are
not measured; a suboptimal controller [36 ] is therefore
To design a control system, employing mainly LQ con-
designed.
trol theory or LQ Gaussian control theory, against the
background of the promotion of research and develop-
4.3.2 Simulations
ment of active dynamic absorbers, the development of
computer technology for realizing the LQ theory cannot Numerical simulations on the discrete model are per-
be ignored. In particular, it is of great significance that formed using the MATLAB Control System Toolbox
the Riccati equation for determining the state feedback routines. In these simulations the values of the com-
control law can be easily solved on a personal computer. ponents that weight velocity terms in Q are all taken
After noting problems in regard to LQ control theory,
this section aims to reveal vibration control performance
when controlled using LQ control theory by creating an
active dynamic absorber. The following explains the
practical application of LQ control theory using the
lumped-parameter model shown in Fig. 18.

4.3 Vibration control of the bridge tower based on

filtered LQ control [35]
4.3.1 State equation
It is not difficult to represent motion of the lumped system
with active dynamic absorbers, as shown in Fig. 18, in
state space. Using the symbols shown in Fig. 18, a state
vector is defined as
X ={ẋ ẋ ẋ ẋ ẋ ẋ ẋ x x x x x x x }T
c 1 2 3 4 5 6 7 1 2 3 4 5 6 7
(10)
where x and x refer to the relative displacements:
6 7
x =x −x , x =x −x (11) Fig. 19 Block diagram of the filtered optimal controller
6 d3 3 7 d4 4
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
526 K SETO

as unity, and the components weighting displacements 4.4 Robust control theory [35]
are taken as zero, since various simulations and experi-
The LQ control theory is considered less versatile than
ments not mentioned in this paper have determined
the H control theory, which is today replacing LQ
that weighting velocity terms seem more effective than 2
control theory as the dominant robust control theory.
weighting displacement terms or a combination of
Further discussion will address the effectiveness of this
both. The feedback gains are changed by adjusting the
theory as a control system design tool using MATLAB,
weighting matrix R. The proper gains examined in the
and DSP. In particular, an interesting experiment has
simulation will be used in the experiments. Therefore, verified its effectiveness in the prevention of spillover.
the simulation is not only a test of the model but also Using the bridge tower under construction shown in
an indispensable procedure to prepare the feedback gains Fig. 16, the effectiveness of H controller with a hybrid
for the experiments. The values of R used in this paper 2
dynamic absorber is demonstrated.
are 0.001. The H norm of a stable transfer matrix G( jv) is
Figures 20 and 21 show the calculated frequency 2
defined as its largest singular value over the entire
response (compliance) and the impulse response respect- frequency range:
ively, when assuming an excitation at mass point 1 and
picking up at mass point 2. The cut-off frequency of the |G( jv)| = sup {s: [G( jv)]} (13)
2 v
low-pass filters is set to 30 Hz. It can be seen that the
five modes can be well controlled. The experiments have where s: represents the largest singular value of G( jv).
shown quite good robustness and agree well with the With H optimal controller design, a controller is
2
results of the simulation, in respect of both the spillover selected to stabilize the nominal system in a way such
that the H norm of a transfer function, which describes
problems and the structure parameter variations [34]. 2
certain design objectives, is minimized (or becomes smaller
than a specified value). In the present study, the objective
is to design a static state feedback controller based on
the reduced-order model to control the vibration of the
first five modes and to suppress possible instability due
to the neglected higher modes. These objectives can be
achieved by solving a mixed sensitivity minimization
problem in the framework of the H optimal control
2
theory. The block diagram of the H mixed-sensitivity
2
problem is shown in Fig. 22 where A , B , C and D are
c c c c
the reduced-order nominal model matrices. Here y is the
measurement, u is the control input, and z and z are
1 2
the regulated variables. W (s) and W (s) are the two
1 2
weighting function matrices to be selected.
In general, model uncertainties of bridge tower
structures include all the neglected high-frequency modes
which may causes the spillover phenomenon. The upper
Fig. 20 Calculated frequency response of the discrete model bound of the system uncertainties are used to choose the
suitable weighting function. Using these, the feedback
controller is designed that can ensure robust stability
against system uncertainties.
The experimental frequency and the time responses
are shown in Figs 23 and 24 [38]. These responses are
obtained using an FFT analyser by exciting mass point 1

Fig. 21 Calculated impulse response of the discrete model Fig. 22 Block diagram of an augmented-control system

Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 527

Fig. 25 Frequency response with additional mass

Fig. 23 Experimental frequency response

Fig. 26 Impulse response with additional mass

Fig. 24 Experimental time response
The control object was a three-DOF system and its
and observing the response at mass point 2. It is seen dynamic characteristics were identified in the neural
that the first four vibration modes are well suppressed network using white-noise excitation. After carrying out
and also higher modes are not excited which implies that the identification, the synthesis of the controller neural
there is no spillover instability. The fifth mode (which network was performed by the back-propagation method.
corresponds to the crane tower) is also well controlled The usefulness of the method was verified experimentally.
in spite of the unavailability of the state corresponding
to mass point 5.
In practice, the parameters of a bridge tower change 4.6 Other topics [40]
with the progress of construction work. To verify the
Variable-structure control theory, i.e. sliding-mode con-
robustness of the proposed control scheme against the
trol theory, has been researched in Russia where it was
parameter variations, an additional mass of 500 g is
shown to be applicable not only to linear systems but
deliberately attached at each of mass points 1 and 2. As
also to non-linear systems with time deformation. With
a result of the added masses, the natural frequency is
recent dissemination of digital control, this theory is
reduced and the equivalent mass is increased by about
attracting new interest and is being applied to robot
50 per cent at each of mass points 1 and 2. In this
arms, with special attention paid to robust control system
case, the controller gain that is computed at the time of
design. The appearance of variations, non-linear terms
simulation is also used in the control algorithm. The
and unknown parameters characteristic of the sliding
frequency and time responses of the controlled structure
condition increases when the state variable is confined
are shown in Figs 25 and 26. These results confirm the
to a changeover surface.
robustness of the proposed active control scheme against
In cases where the nature of the external turbulence
parameter variations.
is known in advance or is easily anticipated, a com-
bination of feedforward control used to offset anticipated
turbulence and state feedback provides an excellent
4.5 Neural network [39]
means of vibration control that can not be obtained from
Recently, an optimal-vibration control of a multiple- feedback control alone. Many noteworthy studies have
DOF system using a neural network has been reported. been reported in this field.
I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
528 K SETO

5 CONCLUSIONS 9 Nagai, M. and Sawada, Y. Optimum control of active

suspension for a high speed ground vehicle. Trans. Japan
In the past, vibration control was employed post design Soc. Mech. Engrs, 1987, 53(487C ), 1750–1755.
and solely as a countermeasure. More recently, vibration 10 Ormondroyd, J. and Den Hartog, J. P. The theory of
the dynamic vibration absorber. Trans. ASME, 1928,
control devices have been included in structural designs
APM-50(7), 9.
from the outset. The active control device built into the 11 Seto, K. and Takita, Y. Vibration control in multi-
Yokohama Landmark Tower described in this paper is degree-of-freedom system by dynamic absorber (report l ).
an important example of this trend. With this concept In Proceedings of the 1987 ASME Design Technology
of designing incorporated vibration control systems, Conference, 1987, DE-Vol. 4, pp. 169–174 (American
active control can demonstrate its effectiveness; this is Society of Mechanical Engineers, New York).
considered a very favourable trend. 12 Mitsuta, S. and Seto, K. Active vibration control of
This paper has outlined in some detail the general structures arranged in parallel. In Proceedings of the 1st
trend towards active control of vibration. It has further International Conference on Motion and Vibration Control
(MOVIC1), 1992, pp. 146–151.
considered some recently developed vibration control
13 Nishimura, H., Yoshida, K. and Shimogo, T. Optimum
devices according to the selection of the control system active dynamic vibration absorber for multi-degree-of-
employed. The idea of preparing a reduced-order physical freedom system (feedback and feed forward control using
model for application in the vibration control of flexible a Kaiman filter). Int. J. Japan Soc. Mech. Engrs, Ser. C,
structures was also examined. This method was shown 1980, 33(4), 513–521.
to provide optimal performance for both the structure 14 Yoshida, K., Shimogou, T., Suzuki, T., Kageyarna, M. and
and the control system. Traditionally, weighting factors Hashimoto, J. Active vibration control for high-rise build-
have been selected for a control system with an active ings using a dynamic vibration absorber driven by servo-
dynamic absorber using LQ control theory. In this moter. Trans. Japan Soc. Mech. Engrs, 1991, 57(534C ),
paper a process was demonstrated by which an optimal 472–477.
15 Spencer, B. F. and Sain, M. K. Controlling buildings: a new
weighting factor can be selected.
frontier of feedback. IEEE Control System, 1997, 17(6),
Recently problems have occurred whereby the desired 19–35.
performance of the controlled objects has not been 16 Abiru, H., Fujishiro, M., Matsumoto, T., Yamazaki, S.
achieved because of obstacles resulting from vibrations and Nagata, N. Tuned active mass damper installed
generated by motion control. An urgent area of research in the Round-Mark Tower. In Proceedings of the 1st
is the simultaneous control of motion and vibration. International Conference on Motion and Vibration Control
Indeed, if the details presented in this paper have pro- (MOVIC1), 1992, pp. 110–115.
vided some reference towards further research in this 17 Tanida, K., Koike, Y., Mutaguchi, M. and Uno, N. Develop-
rapidly evolving field of active vibration control, then ment of hybrid-type mass damper combining active-type
with passive-type. Trans. Japan Soc. Mech. Engrs, 1991,
the present author will have achieved the purpose of its
57(543C ), 485–490.
publication. 18 Koike, Y. and Tanida, K. Application of V-shaped
hybrid mass damper to high-rise buildings and verification
of damper performance. In Proceedings of Structural
REFERENCES
Engineers World Congress, 1998, paper T198-4.
19 Seto, K. and Matsumoto, Y. Vibration control of multiple
1 Tominari, N., Seto, K. and Kamiyama, K. Study on an connected buildings using active controlled bridges. In
electro-hydraulic servo damper. Preprint Japan Soc. Mech. Proceedings of the 3rd World Conference on Structural
Engrs, 1970, 704(1I ), 23–26. Control, 2003, Vol. 3, pp. 253–261 (John Wiley, New York).
2 Okada, Y. and Nakada, T. Analysis and experiments on an 20 Asano, M., Yamano, Y., Koike, Y. and Akagawa, K.
electro-hydraulic torsional servo vibration damper. Bull. Development of active damping bridges and its application
Japan Soc. Mech. Engrs, 1976, 19(131), 477–484. to triple high-rise buildings. In Proceedings of the 6th
3 Tanaka, N., Miyashita, M. and Tominara, N. Output feed- International Conference on Motion and Vibration Control
back control of servodamper system. Bull. Japan Soc. (MOVIC6), 2002, pp. 7–12.
Mech. Engrs, 1976, 20(137), 1278–1284. 21 Fukuda, T. Control of flexible robot arms. Trans. Japan
4 Seto, K. Trends in vibration control. J. Soc. Instrum. Soc. Mech. Engrs, 1985, 51(468C ), 2140–2145.
Control Engrs, 1989, 28(8), 719–726. 22 Nagai, M. and Sawada, Y. Active suspension control for
5 Seto, K. Active vibration control in machinery. J. Acoust. the flexible structure of an elastic vehicle body. Trans. Japan
Soc. Japan, 1991, 12(6), 263–272. Soc. Mech. Engrs, 1987, 53(492C ), 1750–1755.
6 Nagamatu, A. Modal Analysis, 1985 (Baifukan, Tokyo, 23 Yoshida, K., Shimogo, T. and Nishimura, H. Optimal con-
Japan). trol of random vibration by the use of an active dynamic
7 Seto, K. and Suzuki, K. Active control and identification vibration absorber. Int. J. Japan Soc. Mech. Engrs, Ser. C,
technique in vibration problem. J. Japan Soc. Mech. Engrs, 1988, 31(2), 387–384.
1986, 89(811), 635–642. 24 Seto, K., Takita, Y. and Nagamatsu, A. Vibration control
8 Iwata, Y., Ohshio, S. and Suzuki, N. Active control of pre- in a tower structure by a two-dimensional active mass
cision vibration isolation system. Trans. Japan Soc. Mech. damper with an LQ control system. Trans. Japan Soc.
Engrs, 1991, 57(53–4C ), 521–526. Mech. Engrs, 1988, 54(501C ), 1084–1089.
Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering I08803 © IMechE 2004

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015
 CONTROL OF VIBRATION IN CIVIL STRUCTURES 529

25 Nonami, K., Wang, J. W., Sampei, M. and Mita, T. 33 Fuller, C. R., Elliott, S. J. and Nelson, P. A., Active Control
Active vibration control of flexible rotor using H control of Vibration, 1996 (Academic Press, London).
2
theory. In Rotating Machinery and Vehicle Dynamics, 1991, 34 Seto, K. Lumped modeling and robust control for bridge
DE-Vol. 35, pp. 85–91 (American Society of Mechanical towers under construction. In Advances in Structural Control,
Engineers, New York). 1999, pp. 150–196 (International Center for Numerical
26 Allaire, P. E. Magnetic bearing/damper effects on unbalance Methods in Engineering (CIMNE), Barcelona, Spain).
response of flexible rotors. In Proceedings of the AIAA 35 Seto, K., Doi, F. and Ren, M. Modeling and active vibration
Conference, 1987, Vol. 10–14, p. 22. control for bridge tower structures under construction. In
27 Yoshida, K. Sub-optimal control taking account of spillover Proceedings of the ASME Design Engineering Technical
for elastic structure systems. In Proceedings of the IMACS– Conference (DETC’97), 1997, paper DETC97VIB3808.
IFAC International Symposium on Modeling and Simulation 36 Kosut, R. L. Suboptimal control of liner time invariant
of Distributed Parameter Systems, 1987, pp. 285–291. systems subject to control structure constraints. IEEE
28 Balas, M. J. Theory for Distributed Parameter Systems, in Trans. Autom. Control, 1970, AC15-5, 557–563.
37 Dorato, P., Fortuna, L. and Muscato, G. Robust Control
Control and Dynamic Systems, 1982, pp. 361–421 (Academic
for Unstructured Perturbations—An Introduction, Lecture
Press, New York).
Notes in Control and Information Sciences, Vol. 168, 1992,
29 Balas, M. J. Direct velocity feedback control of large space
p. 19 (Springer-Verlag, New York).
structures. J. Guidance, Control Dynamics, 1979, 2(3),
38 Kar, I. N., Seto, K. and Doi, F. Design of H-infinity based
252–253. robust control law to suppress the vibration of a flexible
30 Seto, K. and Mitsuta, S. A new method for making a reduced bridge tower structure. In Proceedings of the 4th Inter-
order model of flexible structures using unobservability and national Conference on Motion and Vibration Control
uncontrollability and its application in vibration control. (MOVIC4), 1998, Vol. 2, pp. 429–436.
In Proceedings of the 1st International Conference on 39 Yoshida, Y. and Kobayashi, T. Application of neural net-
Motion and Vibration Control (MOVIC1), 1992, pp. 152–158. work to the optimal control of active dynamic vibration
31 Seto, K., Ren, M. and Doi, F. Feedback vibration control absorber. In Proceedings of the 1st International Con-
of a flexible plate at audio frequencies by using a physical ference on Motion and Vibration Control (MOVIC1), 1992,
state space approach. J. Acoust. Soc. Am., 1998, 103, Vol. 2, pp. 947–951.
924–934. 40 Iwata, Y., Nakada, S. and Watanabe, A. Sliding mode con-
32 Merovitch, L. Dynamics and Control of Structures, 1990 trol of flexible arm. Trans. Japan Soc. Mech. Engrs, 1991,
(John Wiley, New York). 57(534C ), 536–567.

I08803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part I: J. Systems and Control Engineering

Downloaded from pii.sagepub.com at UNIV CALIFORNIA SANTA BARBARA on July 10, 2015