Engineering Structures 29 (2007) 3390–3403

www.elsevier.com/locate/engstruct

Dynamic analysis of space structures with multiple tuned mass dampers

Y.Q. Guo, W.Q. Chen ∗
Department of Civil Engineering, Zhejiang University, Hangzhou 310027, PR China

Received 7 February 2007; received in revised form 25 April 2007; accepted 8 September 2007
Available online 22 October 2007

Abstract

Formulations of the reverberation matrix method (RMM) are presented for the dynamic analysis of space structures with multiple tuned mass
dampers (MTMD). The theory of generalized inverse matrices is then employed to obtain the frequency response of structures with and without
damping, enabling a uniform treatment at any frequency, including the resonant frequency. For transient responses, the Neumann series expansion
technique as suggested in RMM is found to be confined to the prediction of accurate response at an early time. The artificial damping technique
is employed here to evaluate the medium and long time response of structures. The free vibration, frequency response, and transient response of
structures with MTMD are investigated by the proposed method through several examples. Numerical results indicate that the use of MTMD can
effectively alter the distribution of natural frequencies as well as reduce the frequency/transient responses of the structure. The high accuracy,
lower computational cost, and uniformity of formulation of RMM are highlighted.
c 2007 Elsevier Ltd. All rights reserved.

Keywords: Reverberation matrix method; Space structure; Multiple tuned mass dampers; Vibration control; Frequency response; Transient response

1. Introduction engineering applications of MTMDs [13–15], active control

scheme with MTMDs [16–18], etc. However, the models of
With the increasing demands for the safety, reliability, structures used in these studies are all of a discrete nature,
durability, and serviceability of structures, the theory of of either single-degree-of-freedom (SDOF) or multiple-degree-
vibration control has attained more and more attention in of-freedom (MDOF), solved either by the method of mode
civil engineering with a lot of practices being implemented superposition or by the finite element method (FEM). Lin
on modern infrastructures such as high-rise buildings, long- et al. [19] summarized several commonly used dynamic
span suspension bridges, and offshore platforms. The concept analysis techniques and developed a convenient and precise
of tuned mass damper (TMD) traces back to the 1940s, with method for predicting the dynamic behavior of structures with
the main idea being that by attaching onto the main structure TMD devices, which was also confined to MDOF systems
a secondary mass with properly tuned spring and damper, the as well. In recent years, a few researchers began to use the
structural vibration near the natural frequency of the structure distributed-parameter model in the dynamic analysis of simple
to which the TMD is tuned can be effectively reduced, but other structures (simply supported beams or continuous beams)
contiguous frequencies may be marginally suppressed or even with TMDs or MTMDs. For example, Chen and Chen [20]
amplified. As a solution, Clark proposed in 1988 to use multiple investigated the dynamic response of a single-span girder with
tuned mass dampers (MTMD) to control the structural vibration TMDs under moving loads based on a Timoshenko beam
in a wider frequency band, using more than one TMD and model. Lin et al. [21] and Li et al. [22] studied the vibration
with each tuned to a different dominant frequency [1]. Since control of bridge girders with MTMDs under high-speed train
then, a lot of papers on the dynamic behavior of structures loads using an Euler–Bernoulli beam model, which was solved
with MTMD have been published [2–6]. Research interests by the method of mode superposition.
also include the optimization of MTMD parameters [7–12], A generalization of the reverberation matrix method
(RMM) [23–26], which was based on the concept of elastic
∗ Corresponding author. Tel.: +86 571 87952284; fax: +86 571 87952165. wave propagation, is presented to investigate the dynamic
E-mail address: [email protected] (W.Q. Chen). response of a three-dimensional framed structure with MTMDs.

0141-0296/$ - see front matter c 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2007.09.004
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3391

(a) A collection of typical members of (and the coordinate systems for) the (b) Sign convention of physical variables.
structure with MTMD.

Fig. 1. A typical part of structures with MTMD.

The structural members are modeled as a classical longitudinal principal axes of inertia of the cross section of the member.
rod for the axial wave motion, a Timoshenko beam for flexural The directions of y J K and y K J are opposite to each other, while
waves, and a classical torsional shaft for the torsional mode. those of z J K and z K J are identical, following the right-handed
A small deformation is assumed so that the linear theory is screw rule. The coordinate systems and the sign convention of
valid and all three types of wave motions are uncoupled from physical variables are depicted in Fig. 1.
each other for a single structural member. The effect of viscous To make it clear, any physical variable pertaining to the local
damping, which was not considered in the original version coordinate system (x J K , y J K , z J K ) will be indicated by the
of RMM [23,24,26], is included as an extension. A uniform superscripts JK. A hat will be used to denote the quantities
treatment of joints of structures is suggested to keep the number in the frequency domain (the transformed quantities). For an
of unknowns in the formulation of the RMM for structures with arbitrary member J K , the equations of motion governing
MTMDs the same as that without MTMDs. It is noted that the the wave propagation in a one-dimensional space member is
current formulation is presented in a way that is particularly (superscripts JK omitted)
convenient for programming. Furthermore, the frequency
∂ 2u ∂ 2u ∂u
response and the long-time transient response of structures EA = ρ A + ηu A (1)
are obtained by the generalized inverse matrix and artificial ∂x2 ∂t 2 ∂t
∂ v ∂φz ∂ 2v ∂v
 2 
damping technique, respectively, in contrast to the Neumann
κG A − = ρ A 2 + ηv A
series expansion as adopted in Refs. [23–26]. Numerical ∂x 2 ∂x ∂t ∂t
examples are considered to validate the proposed method, and ∂ w ∂φ y
 2 
∂ w
2 ∂w
the influence of MTMD on the natural frequencies, frequency κG A − = ρ A 2 + ηw A
∂x 2 ∂x ∂t ∂t
responses, and transient responses of the structures is discussed.
∂ 2 φx ∂φx ∂ 2 φx
G Ix = η φx I x + ρ I x
2. Analytical model of structures with MTMD ∂x2 ∂t ∂t 2
∂ φy
2 ∂w ∂ 2φy ∂φ y
 
A typical collection of structural members with MTMDs E Iy + κG A − φ y = ρ I y + ηφy I y
∂x 2 ∂x ∂t 2 ∂t
is shown in Fig. 1. The laying points of the TMDs, the
∂ 2 φz ∂v ∂ 2 φz ∂φz
 
acting points of loads, and the connecting points of structural E Iz + κG A − φz = ρ Iz 2 + ηφz Iz
members are all treated as joints. Any material segment ∂x2 ∂x ∂t ∂t
between two end joints, say joints J and K , then constitutes where u(x, t), v(x, t) and w(x, t) are displacements in the
a structural member JK in the analytical model, the carrier of directions x, y and z, respectively; φx (x, t), φ y (x, t) and
longitudinal, flexural, and torsional wave motions for which φz (x, t) are rotations about the x-, y- and z-axes, respectively;
the classical theory of rod, Timoshenko beam theory, and the Young’s modulus E, shear modulus G, shear coefficient κ,
the classical theory of torsion will be employed, respectively. mass density ρ, cross-sectional area A, and the moments of
Let us set up a global coordinate system (X, Y, Z ) for the inertia of the cross section about x-, y- and z-axes I x , I y and
whole structure and a dual set of local coordinate systems Iz , are all assumed to be constant for each member; ηu , ηv , ηw ,
(x J K , y J K , z J K ) and (x K J , y K J , z K J ) for any member J K ηφx , ηφy and ηφz are viscous damping coefficients, which can
with x J K directing from J to K and x K J from K to be expressed as ηu = ηv = ηw = ηφx = ηφy = ηφz = 2ξρω
J . The axes y J K , z J K , y K J and z K J coincide with the if they are frequency dependent and ηu = ηv = ηw = ηφx =
3392 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

ηφy = ηφz = 2ξρ if they are frequency independent, here ξ UmJ (t), which have components as follows
is a constant damping ratio, assumed to be the same for all six
generalized displacements for simplicity. P J (t) = [PXJ (t), PYJ (t), PZJ (t), m XJ (t), m YJ (t), m ZJ (t)]T (8)
Denote the resultant force and displacement vectors at U (t) = [U XJ (t), UYJ (t), U ZJ (t), Φ XJ (t), ΦYJ (t), Φ ZJ (t)]T
J
(9)
x J K in the local coordinate system (x, y,
z) J K in the time
J K x , t and δ J K x J K , t respectively, and
J K UmJ (t) = [UmJ X (t), UmY J
(t), UmJ Z (t),


domain as F
JK Φm X (t), ΦmY (t), ΦmJ Z (t)]T .
J J
(10)
x J K , ω and


their counterparts in the frequency domain, F̂
JK
δ̂ x J K , ω , are


3.1. Equations of motion for the TMD mass
JK
F̂ = [ N̂ J K , Q̂ yJ K , Q̂ zJ K , M̂xJ K , M̂ yJ K , M̂zJ K ]T , (2) Applying Newton’s law to the TMD mass in the global
JK
coordinate system gives
δ̂ = [û J K , v̂ J K , ŵ J K , φ̂xJ K , φ̂ yJ K , φ̂zJ K ]T . (3) J
h
J J
i
M J Üm (t) + C J U̇m (t) − U̇ (t)
The pair of Fourier transforms that relate a quantity in the time h i
domain to that in the frequency domain are defined as + K J UmJ (t) − U J (t) = 0 (11)
Z ∞
fˆ(ω) = f (t)e−iωt dt, where the mass, damping, and stiffness matrices of the TMD
−∞ read as
Z ∞ (4)
1
f (t) = fˆ(ω)eiωt dω. M J = diag[m J , m J , m J , m J (r XJ )2 , m J (rYJ )2 , m J (r ZJ )2 ] (12)
2π −∞
C J = diag[c XJ , cYJ , c ZJ , cΦ
J
X , cΦ Y , cΦ Z ]
J J
(13)
Applying the above Fourier transform to the governing
equations in Eq. (1) gives a set of ordinary differential equations
J
K = diag[k XJ , kYJ , k ZJ , kΦ
J
X , kΦ Y , kΦ Z ]
J J
(14)
in one spatial coordinate, whose solution can be easily obtained
in which m J is the mass of the TMD device attached to joint
as
J ; r XJ , rYJ and r ZJ are the radii of gyration of the TMD mass
JK
about its respective axes; c XJ , cYJ , c ZJ , cΦ J , cJ J
   
F̂ = A Jf K x J K , ω a J K (ω) + D Jf K x J K , ω d J K (ω) (5) X Φ Y and cΦ Z are
the viscous coefficients of the TMD damper and k X , kYJ , k ZJ , J
JK
   
δ̂ = AuJ K x J K , ω a J K (ω) + DuJ K x J K , ω d J K (ω) (6) kΦJ , k J and k J are the elastic constants of the TMD spring.
X ΦY ΦZ
Note that rotational stiffness and damping are included in the
where above model to make the analysis general. In a usual manner,
the damping matrix can be assumed to be
a J K (ω) = [a1J K (ω) , a2J K (ω), a3J K (ω) , q q q q
a4J K (ω), a5J K (ω), a6J K (ω)]T C J = 2ξT diag m J k XJ , m J kYJ , m J k ZJ , m J (r XJ )2 kΦJ ,
X
(7)
d J K (ω) = [d1J K (ω) , d2J K (ω), d3J K (ω) , q q 
d4J K (ω), d5J K (ω), d6J K (ω)]T m J (rYJ )2 kΦ
J , m J (r J )2 k J
Y Z ΦZ

with a Jj K and d jJ K ( j = 1, 2, . . . , 6) being the arriving where ξT is a constant damping ratio. For simplicity, an identi-
wave amplitudes and departing wave amplitudes of various cal value of ξT has been assigned to all six degrees-of-freedom
waves in member J K , respectively. a J K and d J K are termed of the TMD mass. Nevertheless, different values of ξT can be
accordingly as the arriving wave
vector and departing adopted and the formulations should be modified accordingly.

wave
vector, respectively. A Jf K x J K , ω and D Jf K x J K , ω are the Applying the Fourier transform to Eq. (11), we obtain
resultant force phase matrices for the arriving and departing
−1 
J
  J
waves, respectively, while AuJ K x J K , ω and DuJ K x J K , ω Ûm (t) = −ω2 M J + iωC J + K J iωC J + K J Û (15)

are the displacement phase matrices. The expressions for these

phase matrices are given in Appendix A. which expresses the displacement of TMD mass in terms of the
joint displacement in the frequency domain. It should be noted
that Eq. (11) should be replaced by equilibrium equations about
3. Formulations of the reverberation matrix method
the translational displacements only when the rotatory inertia of
the TMD mass is neglected. In this case, the mass, damping and
To establish the formulations of the RMM, we need to stiffness matrices will be degenerated into diagonal matrices of
specify the conditions of joints. In Fig. 1(a), joints J and M order 3, but the form of Eq. (15) is unchanged.
are rigidly attached with a TMD, joint N is pin-connected, also
with a TMD, while others are those without a TMD. Taking 3.2. Equilibrium conditions of a joint with a TMD
joint J for example, denotes the external force vector and
the displacement vector of joint and TMD mass in the global If the number of members connected to a joint J is n J , then
coordinate system (X, Y, Z ) respectively as P J (t), U J (t) and the equilibrium conditions of the joint are given by
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3393
h i
nJ h i JK
AuG (0, ω) a J K (ω) + DuG
JK
(0, ω) d J K (ω)
X J J
(T J K )T F J K (0, t) + C J U̇m (t) − U̇ (t)
K =1 J
h i = Û (ω), (K = 1, 2, . . . , n J ) (23)
+K J
UmJ (t) − U J (t) + P (t) = 0
J
(16) where
where T J K is the coordinate transformation matrix between
JK
AuG (0, ω) = (T J K )T AuJ K (0, ω) ,
(24)
the local coordinate system (x J K , y J K , z J K ) and the global JK
DuG (0, ω) = (T J K )T DuJ K (0, ω)
coordinate system (X, Y, Z ). The elements of T J K can be given
in terms of the global coordinates of the two end joints of the are the phase matrices of displacements for arriving waves and
member JK, as shown in Appendix B. Here, the mass of joint departing waves respectively in the global coordinate system.
J is neglected, and if not, it can be easily incorporated into the Eq. (23) can be expressed in a matrix form by incorporating the
formulations without any difficulty [24]. joint wave vectors as
Using the Fourier transformation and using Eq. (15), one J
A2J a J + D2J d J = U1J Û (25)
obtains
nJ
where
JK T JK J
X
(T ) F̂ (0, ω) + P̂ J1
A2J = diag[AuG (0, ω), AuG J2
(0, ω), . . . ,
K =1 JnJ
−1 AuG (0, ω), . . . , AuG (0, ω)]6n J ×6n J
JK
  (26)
= −ω2 M J −ω2 M J + iωC J + K J J1
D2J = diag[DuG (0, ω), DuG J2
(0, ω), . . . ,
JnJ
  J DuG (0, ω), . . . , DuG
JK
(0, ω)]6n J ×6n J
× iωC J + K J Û h i
J
U1J J = [I6×6 , I6×6 , . . . , I6×6 ]T . (27)
= K1J Û (17) 6n ×6

where K1J signifies the equivalent stiffness matrix due to the 3.4. Scattering of elastic waves at a joint with a TMD
TMD device at joint J . With the phase matrices of the resultant
force vector in the global coordinate system defined as For a practical problem, among the twelve quantities of a
joint (six displacements and six forces), there are always six
G (0, ω) = (T ) A f (0, ω),
JK T JK
A Jf K knowns and six unknowns. Denote the unknown quantities as
(18) J J
Û E and the known quantities as P̂ E . Combining Eqs. (20) and
G (0, ω) = (T ) D f (0, ω),
JK T JK
D Jf K
(25), we get
we obtain from Eqs. (5), (17) and (18) ! ! ! 
A1J D1J K E1
J
dJ J

J K E2 J
nJ h a + J = P̂ E (28)
A2J D2J T1 Û E T2
X i
G (0, ω) a
A Jf K (ω) + D Jf K
G (0, ω) d (ω)
JK JK

K =1 where K E1J and K E2 J are diagonal matrices with their

J J components being either +1 or −1, or equal to the
+ P̂ (ω) = K1J Û (ω). (19)
corresponding components in +K1J or −K1J ; T1 and T2 are also
By assembling the arriving and departing wave vectors for each diagonal matrices with components being either +1 or −1. The
connected members at joint J into joint wave vectors a J (ω) and concrete form of these matrices is determined according to the
d J (ω), Eq. (19) can be written in a matrix form as prescribed quantities of the joint.
The local scattering relations can be written in matrix form
J J
A1J a J + D1J d J + P̂ = K1J Û (20) of d J = S J a J + s J [23–26], with the scattering matrix S J and
the source vector s J given by
where
!−1 !
J J J
h
A1J = A Jf 1G (0, ω) · · ·
J
i
D K A
G (0, ω)
A Jf K A Jf nG (0, ω) 1 E1 1
S J = − I6n J ×6n J 06n J ×6


··· (29)
6×6n J
h i (21) D2J T1 A2J
D1J = D Jf 1G (0, ω) · · · G (0, ω)
D Jf K ··· JnJ
D f G (0, ω) J
.
6×6n !−1 

D1J K E1 J J

K E2 J
J
s = I6n J ×6n J 06n J ×6 J
P̂ E . (30)
D2 T1 T2
3.3. Compatibility conditions of joint with a TMD
If all displacements of the joint are known, the scattering matrix
The displacements of all connected elements at joint J and the source vector can be directly obtained from Eq. (25)
should be compatible with that of joint J according to  −1  −1 J
S J = − D2J A2J , s J = D2J U1J Û . (31)
(T J K )T δ J K (0, t) = U J (t), (K = 1, 2, . . . , n J ). (22)
Eq. (20) is then used to compute the reaction forces after
This gives, with the help of the Fourier transform and Eq. (15), the wave vectors are obtained. Note that the elements of the
3394 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

local scattering matrix S J represent appropriate reflection or phase matrix that is block-diagonal and d̄ the global departing
transmission coefficients of waves arriving at the joint J . wave vector that has the same elements of d but sequenced in
For the pin-connected joint with a TMD, the scattering a different order. The two are related by d̄ = U P d, where U P
relations are derived similarly, with the compatibility condition is the permutation matrix of 12m × 12m with only one element
about rotation replaced by the condition of a zero moment at the in each row and column being unit and the others being zero.
joint. Meanwhile, the equilibrium equation about the moment Thus, the global phase relation finally takes the form
becomes meaningless. In this case, the rotation of the joint has
no influence on the dynamics of the structure, and hence both a(ω) = P (ω) U P d(ω). (35)
numbers of independent external forces and joint displacements
that shall be considered in the analysis decrease from six to 3.7. The reverberation matrix and complex frequency-response
three. function matrix
The local scattering relations of the joint without a TMD can
be obtained in the same way. A detailed account can be found Combining Eqs. (32) and (35) gives
in Ref. [24]. [I − R(ω)]d(ω) = s(ω) or d(ω) = H(ω)s(ω) (36)
3.5. Global scattering relations of structures with an MTMD where R(ω) = S (ω) P(ω)U P is called the reverberation
(ray) matrix [23–26], and H(ω) = [I − R(ω)]−1 . Since Eq.
The global scattering relations can be formed by simply
(36) express the relations between amplitudes of departing
assembling the local scattering relations at all joints. In matrix
waves from all joints and the external actions at the joints of
form, it is expressed as
the structure, H(ω) is called the complex frequency-response
d = Sa + s (32) function matrix, according to the definition of the complex
frequency-response function of a SDOF system [28].
where S is the global scattering matrix of 12m × 12m and s is
the global source vector of 12m × 1 given by 4. Dynamic analysis of structures with MTMD
h i
S = diag S1 , S2 , . . . , S J , . . . , Sn , 4.1. Free vibration analysis
h iT (33)
s = s1 , s2 , . . . , s J , . . . , sn If there is no external force applied at the joints of the
structure, then s(ω) = 0 in Eq. (36). To have a non-trivial
with m and n being, respectively, the total number of members solution, the determinant of the coefficient matrix (the inverse
and joints of the whole structure. of the complex frequency-response function matrix) should
vanish, giving the following frequency equation of the structure
3.6. Phase relations
|I − R(ω)| = 0. (37)
The wave arriving at joint J in the local coordinate system
(x, y, z) J K is exactly the same as the corresponding departing The roots of this equation, i.e. the natural frequencies of the
wave from joint K in the local coordinate system (x, y, z) K J , structure, can be searched by a bisection method applied to
except for a phase shift for the wave traversing the member. both the real and imaginary parts of the determinant. The
Thus, we arrive at the following phase relations for the member corresponding vibration mode is then determined by first
JK calculating the characteristic vector of d from Eq. (36) with
  the zero right-hand side, then calculating the corresponding
a J K (ω) = P J K l J K , ω d K J (34) arriving wave vector a from Eq. (35), and finally obtaining the
displacements from Eq. (6).
where a J K (ω) = [a1J K , a2J K , . . . , a6J K ]T , and d K J (ω) =
[d1K J , d2K J , . . . , d6K J ]T ; P J K l J K , ω is called the local phase


4.2. Frequency response analysis
matrix given by
  For undamped structures, the complex frequency-response
JK JK JK
P J K l J K , ω = diag[−e−ik1 l , −e−ik2 l , −e−ik3 l , function matrix H(ω) has an infinite number of poles along the
JK JK JK
real axis of the complex ω-plane. Obviously, at these poles,
− e−ik4 l , −e−ik5 l , −e−ik6 l ]. the wave amplitudes become infinite, while near these poles,
From Appendix A, it is known that for all kinds of waves, inaccurate results usually will be obtained based on the direct
if k is the wave number, then −k is also the wave number. The inverse of I − R(ω) due to the singular behavior of H(ω)
wave number ki (i = 1, 2, . . . , 6) in the phase matrix may be at the poles. In the transient response analysis, Howard and
chosen to guarantee Re[−iki l J K ] ≤ 0. Thus, no large number Pao [23] and Pao et al. [24] suggested using the Neumann series
is included in the phase relations and the numerical instability expansion,
encountered in the transfer matrix method (TMM) [27] will be
[I − R(ω)]−1 = I + R(ω) + R2 (ω) + · · · + R N (ω) + · · · (38)
avoided completely in the RMM.
All local phase relations can be assembled into a global to avoid the singular behavior at poles in the Fourier inverse
phase relation in matrix form of a = Pd̄, with P being the global transformation. The necessary and sufficient conditions for the
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3395

Neumann series to be absolutely convergent is that the spectral truncated Neumann series should be specified as the scattering
radius of matrix R is less than unit [29]. If we use this technique times (N ) of the fastest wave passing the shortest path from
to calculate the frequency response, we find quite different the source to the observation point according to the observation
results will be obtained when a different number of total terms time. To minimize the aliasing error, the time period in FFT
is adopted in the Neumann series. This will be illustrated should be set as the sum of the time of observation and
through a numerical example in the next section. the time needed by the slowest wave to pass through the
To obtain the frequency response near the poles as accurately longest path with N times of scattering [24]. However, since
as that far from the poles, the theory of generalized inverse the flexural wave is dispersive in that it is very slow for a
matrices (GIM) [29] is employed, with the definition of lower frequency, it becomes difficult to specify the time period
the Moore–Penrose GIM given in Appendix C. Since the properly. In the computation, a longer period is then needed
generalized inverse matrix is obtained in the sense of least to include the contribution of waves with low frequencies.
square approximation, the frequency response at the poles For medium and long time transient responses, since waves
can also be treated in an identical manner without the need scatter many times along with wave mode conversion, tracing
to specify the program, although with a finite value of waves and determining N accurately becomes very difficult for
the calculated departing wave vector that is actually more a complex structure. The dispersive nature of a flexure wave
physically realistic because of the damping in real structures. becomes more critical and imposes a huge obstacle on the
The validity of GIM will be demonstrated by the numerical determination of N . In fact, as will be shown in the numerical
example in the next section. example, results for different N ’s will differ obviously from
The frequency response of various physical variables at x J K each other. Furthermore, a longer time period should be used in
of member J K in the local coordinate system (x, y, z) J K may FFT to reduce the aliasing error due to the dispersive flexural
be computed by waves. However, the time interval should be kept small to
JK meet the requirement of cut-off frequency. Thus, the numerical
R̂i (x J K , ω) = (AiJ K (x J K , ω)EPU P efficiency also becomes undesirably low. So the Neumann
+ DiJ K (x J K , ω)E) d(ω) (39) series expansion technique is unsuitable for the calculation of
medium and long time transient responses.
JK
where R̂i can be a vector of strain, stress, resultant force, To overcome this difficulty, a direct method is suggested
displacement, velocity and acceleration, E is a rectangular here. For undamped and lightly damped structures, the artificial
matrix serving to extract the local wave vectors a J K (d J K ) damping technique (ADT) [30–32] that has been validated
from the global wave vectors a (d), and AiJ K and DiJ K are the for single and multiple-degrees-of-freedom is employed here
corresponding local phase matrices of the arriving waves and for continuous structural systems. By adding appropriate
departing waves respectively, which are easily derived from the viscous damping (frequency independent) into the model, the
phase matrices of displacements given in Appendix A. singularity or almost singular behavior near the real axis
of the complex ω plane of the complex frequency response
4.3. Transient response analysis function is removed, and the aliasing and overlapping error
can be minimized significantly. Hence precise results could
The transient response is obtained by the inverse Fourier be anticipated. It is noted that the damping is introduced here
Transform of the frequency response as directly into the physical model by involving the artificial
1
Z +∞ viscous force, rather than by adding a small imaginary part
JK
RiJ K (x J K , t) = R̂i (x J K , ω)eiωt dω to the frequency parameter in the inverse Fourier transform as
2π −∞
Z +∞ shown in Refs. [30–32] for discrete systems. The effectiveness
1 of ADT for the dynamic analysis of continuous structural
= (AiJ K (x J K , ω)EPU P
2π −∞ systems in the frequency domain will be testified soon through
numerical examples. For moderately and heavily damped
+ DiJ K (x J K , ω)E)(I − R)−1 seiωt dω. (40)
structures, accurate results could be obtained without the
The integral can be evaluated numerically via the FFT necessity of introducing any additional damping, because the
algorithm. For transient response at an early time, since the poles of the complex frequency-response function matrix are
scattering times of the waves is relatively small and the far from the real axis of the complex ω plane and the effect of
dispersive characteristic of flexural waves has little influence aliasing and overlapping is negligible.
on the tracing of wave propagation in the structure, (I − R)−1 Thus, the transient dynamic response of all structures
can be expanded into the Neumann series as in Eq. (38). This including undamped, lightly damped, moderately damped,
can effectively avoid the singularities encountered in the inverse and heavily damped systems can be evaluated uniformly by
transform, which is generally treated by the residual theorem the present direct method with the formulations of RMM,
in theoretical analysis but is very difficult to be implemented regardless of the time of interest (early time, medium period,
in the numerical calculation [24]. Furthermore, each term in or long duration). However, since the RMM with the Neumann
the Neumann series corresponds to the contribution of waves series expansion can predict more accurate response at an
with appropriate scattering times, giving a clear physical image early time, and also bears a clear physical image of wave
of wave propagation in the structure. The highest power in the propagation in the structure, it is still preferable to use the
3396 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

Table 2
First ten frequencies for the continuous beam without MTMD (rad/s)

Order FEM-10 FEM-200 FEM-1000 RMM

1 3.1885 3.1870 3.1870 3.1869
2 5.2226 5.2162 5.2162 5.2158
Fig. 2. The three-span continuous beam. 3 6.2678 6.2568 6.2568 6.2559
4 12.0612 11.9877 11.9877 11.9850
5 18.5517 18.3035 18.3035 18.2990
Table 1
6 20.2595 19.9485 19.9485 19.9420
Parameters of the continuous beam
7 26.8418 26.2336 26.2336 26.2220
Young’s Shear Shear co- Mass Area of Moment 8 42.7081 38.4342 38.4342 38.4160
modulus modulus efficient density ρ cross- of 9 46.8185 41.8617 41.8617 41.8360
E (Pa) G (Pa) κ (kg/m3 ) section A inertia 10 53.7947 46.7425 46.7425 46.7060
(m2 ) Iz (m4 )
3.25 × 1010 1.354 × 1010 0.8434 2500 13.6 0.1592 Next, assume that a harmonic load of amplitude 1000 kN and
frequency varying in the range of 0–400 rad/s is applied at the
conventional RMM for calculating the early-time transient mid-point of the middle span. The bending strains of the outer
response (scattering times N < 30) [24]. It is also noted that fiber of the mid-section of the left span calculated by the FEM
although divergence will occur for longer time responses of and different schemes of the RMM are given and compared in
undamped structures by ADT, it is actually more physically Fig. 3.
realistic, because damping always exists in a real structure. It is seen from Fig. 3(a) that the locations of peaks predicted
by FEM-1000, RMM with the generalized inverse matrix
5. Numerical examples (RMM-GIM), and RMM with the Neumann series expansion
(RMM-1000, here and after the numeral indicates the number
5.1. Continuous beams of terms in the truncated series) agree well, but the amplitude
of RMM-1000 does not coincide with the other two, not even at
At first, a continuous beam without an MTMD depicted in the poles of the complex frequency-response function matrix.
Fig. 2 is considered to testify to the validity of the proposed In fact, both curves for FEM-1000 and RMM-GIM are quite
method. The material and geometric parameters used in the smooth, while that for RMM-1000 is very rough as shown
computation are given in Table 1. clearly in the insert of Fig. 3(a). Although not shown here, more
The first ten frequencies of the continuous beam are given in results indicate the good performance of RMM-GIM, except
Table 2, and are compared to the FEM calculations. In the table, a difference from the large-scale FEM at the poles, as also
FEM-1000, for instance, denotes that the beam is discretized shown in Fig. 3(a), where the physical implication is already
into 1000 elements of equal length and the commercial software clear. Also, the numerical efficiency of RMM-GIM is higher
ANSYS is utilized. than that of FEM with a fine mesh, to be shown later. Hence,
It is seen that the FEM results approach those of the RMM the RMM combined with the generalized matrix technique is
with the increasing of element numbers. The results of FEM- more suitable for the frequency response analysis of structures
200 and FEM-1000 are very close to each other and are always without damping.
larger than those of the RMM, because the stiffness of the FEM Fig. 3(b) shows that, using different terms in the Neumann
beam model is larger than that of the actual continuous beam. series usually gives quite different results, further indicating

(a) Comparison between the FEM, RMM-GIM, and RMM-1000. (b) Comparison between the RMM-GIM and RMM with different terms of
Neumann series.

Fig. 3. Frequency responses of bending strain of the mid-section of member 1–2.

 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3397

Table 3
Comparison of computational efficiency between RMM and FEM

Method Stored data Computational time (s)

Natural Frequency Transient Transient
frequency response response response
0.5 s 10 s
ANSYS 12 007 (at 4.000 224.000 400.400 800.400
least)
RMM 2 × 2352 1.766 2.813 17.469 37.078

makes it extremely difficult to properly determine the number

of terms for long time response. The aliasing error also becomes
very significant. Thus the Neumann series expansion technique
could not work as well as that for the early time response, which
Fig. 4. Transient response of the bending strain of the mid-section of members
1–2 (t < 0.2 s).
was shown in Pao et al. [24]. On the other hand, as shown in
these figures, the results of the RMM-ADT agree well with that
of the FEM with a fine mesh even for a long period.
The computational efficiency of the RMM is compared to
that of FEM-1000 for the calculation of the first ten natural
frequencies, the bending strain frequency response and the
transient response at the mid-section of the left span with
a personal computer of 850MIPS, as given in Table 3. The
higher efficiency of the RMM (using the GIM and ADT for
frequency response and transient response, respectively) is
clearly highlighted by this example.
We then turn to consider the beam with three identical
TMDs, each attached to the mid-section of one span of the
continuous beam. It is noted that the locations of the TMDs
could have an important effect on the vibration suppression of
a structure, and an optimization analysis is generally required
Fig. 5. Transient response of the bending strain of the mid-section of member to locate them properly, which is, however, outside the scope
1–2 (t < 0.5 s).
of the current study. The mass ratio of all the TMD masses
that the Neumann series expansion technique is inapplicable over the main structure is set to be µT = 5%. The natural
to the frequency response analysis. From the discussion of frequency of all the TMDs is specified as ω = 3.18 rad/s,
Section 4, it is understandable, since the frequency response which is exactly the same as the first natural frequency of the
(steady-state response) actually corresponds to a very long time beam without MTMD. Hence, the mass and the linear spring
behavior after the harmonic excitation is applied. constant of each TMD are m = 5.67 × 104 kg and k X = kY =
Now consider that at the mid-point of the middle span of k Z = 5.76 × 105 N/m, respectively, and the rotatory inertia
the continuous beam is applied a dynamic load of a rectangular of the TMD masses is neglected. It should be also emphasized
pulse that the above assumption of TMDs with identical parameters is
 just for simplicity. In fact, the natural frequency of each TMD
1000 kN 0 ≤ t ≤ 0.1 s can be arbitrary in our analysis and the dynamic problem of
p(t) = (41)
0 kN t > 0.1 s. structures with MTMDs of various distributed frequencies and
The bending strain responses at the mid-section of the left span parameters can be easily treated in a similar way.
calculated by the FEM, RMM with Neumann series expansion, Using the proposed method, the first several natural
and RMM with artificial damping technique (RMM-ADT) are frequencies (<50 rad/s) of the beam with MTMDs and the
presented in Figs. 4–6 for periods of duration 0.2 s, 0.5 s and transient response of the bending moment at the mid-section
10 s, respectively. In FFT, a time step 0.0005 s for periods 0.2 s of the middle span (joint 3) are calculated and given in Table 4
and 0.5 s and a time step 0.005 s for period 10 s are adopted, and Fig. 7 respectively. In the calculation, the damping ratio of
all with a sampling number 16 384. The artificial damping TMDs is taken to be ξT = 0.02.
(frequency independent) is selected as ξ = 0.36, according to From Table 4, it is seen that there are totally 12 frequencies
the general criteria suggested in Refs. [30,32]. that are less than 50 rad/s for the beam with MTMDs. On the
It is seen that the number of terms in the Neumann series contrary, only 10 frequencies can be obtained for the original
has a significant influence on the results. Precise results are beam (without MTMD). This indicates that an MTMD has
obtained only when a proper number of terms is selected. As a significant effect on the distribution as well as the values
discussed earlier in Section 4, the dispersion of flexural waves of the natural frequency, which can be utilized to adjust the
3398 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

(a) Response for 0 ≤ t ≤ 10 s. (b) Details for 7.7 ≤ t ≤ 9.0 s.

Fig. 6. Transient response of the bending strain of the mid-section of member 1–2 (t < 10 s).

Table 4
Frequencies less than 50 rad/s of the continuous beam (rad/s)

Order Without MTMD With MTMD

1 3.1869 2.7244
2 5.2158 3.0794
3 6.2559 3.6941
4 11.9850 5.3575
5 18.2990 6.3838
6 19.9420 11.9880
7 26.2220 18.3040
8 38.4160 19.9450
9 41.8360 26.2370
10 46.7060 38.4330
11 41.8440
12 46.7240

Fig. 7. Transient response of the bending moment at joint 3 (t < 5.0 s).
dynamic properties of a structure to avoid the resonant response
at specified frequencies. of an MTMD to the beam may introduce extra modes, which
Fig. 7 shows that the transient responses of the beams with can also be seen from Table 4. The shift of natural frequency
and without MTMD are very close to each other before 1.0 s. due to MTMDs from the original one is also clearly observed
In fact, at the very early time stages, the function of the MTMD form Fig. 8(a) and (b). Furthermore, Fig. 8(a) and (b) indicates
has not been fully achieved. However, the response of the beam that the bending moment amplitude does not always decrease
decreases obviously with the attachment of the MTMD after with the damping ratio of the TMD. All these interesting
1.0 s. The longer the time, the more evident the reduction of results suggest that an accurate dynamic analysis should be pre-
response. Since the load is a rectangular pulse of finite duration, conducted to assure the desired effect of MTMD on the control
the beam with MTMD will eventually come to be at rest due to of practical structures.
the damping characteristic of the MTMD.
The frequency responses of the bending moment at the mid- 5.2. Two storey space frame
section of the middle span are first calculated for the beam
without an MTMD, and those near the first four peaks are The second example considered is a two-storey three-
presented in Fig. 8, where the counterparts for the beam with dimensional framed structure with MTMD, as depicted in
MTMDs with different values of the damping ratio are also Fig. 9. The material and geometric parameters adopted in the
given for comparison. Note that the bending moment is deemed computation are listed in Table 5. The damping ratio of the
to be positive when the lower fiber of the cross-section is under frame is assumed to be ξ = 0.005, and the damping coefficient
tension. Abundant phenomena of the effect of the MTMD on is frequency dependent. Parameters of all TMDs are assumed to
the response at different frequencies can be seen from the be identical, with a damping ratio ξT = 0.05 and a natural fre-
figure. While Fig. 8(c) and (d) show that the bending moment quency ω = 27.94 rad/s. The latter value of frequency equals
amplitude near the third and fourth peaks always decreases with to the first natural frequency of the frame without damping.
the damping ratio, more complicated variations are presented Assume that a harmonic load of amplitude 1000 kN and
in Fig. 8(a) and (b). For example, double peaks appear in frequency varying in the range of 0–200 rad/s is applied at joint
Fig. 8(a) for certain values of ξT , indicating that the attachment 15 along the Z -axis. The bending strain response of the outer
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3399

(a) Near the first peak. (b) Near the second peak.

(c) Near the third peak. (d) Near the fourth peak.

Fig. 8. Frequency response of bending moment at joint 3.

Table 5
Parameters of the space frame

Material Young’s Shear Shear Mass density ρ

parameters modulus E (Pa) modulus coefficient (kg/m3 )
G (Pa) κ
2.0 × 1011 7.6923 × 0.8434 7800
1010

Geometric Area of Moment of Moment of Moment of

parameters cross-section inertia inertia inertia Iz (m4 )
A (m2 ) I x (m4 ) I y (m4 )

0.1 × 0.1 1.6667 × 8.3333 × 8.3333 × 10−6

10−5 10−6

the application of an MTMD, since the tuned mass could not be

arbitrary and an optimum value should be determined according
to the analysis based on an accurate model.
Fig. 9. A two-storey space frame with MTMD. The transient response of the frame is also calculated by the
proposed method when the rectangular pulse load in the form
fiber of the mid-section of member 8–9 about the X -axis around of Eq. (40) is applied at joint 15 in the Z -direction. In the FFT,
the first two natural frequencies are given in Fig. 10 for different a time step 0.001 s and a sampling number 16 384 are adopted.
values of µT , the mass ratio between MTMD and the frame. The bending strain response at the mid-section of member 8–9
It can be seen that the mass ratio has a more obvious effect on about the X -axis is given in Fig. 11 for different values of the
the frequency response near the first resonant frequency than on mass ratio µT . It is seen that a better reduction effect at the early
that near the second resonant frequency. When µT > 3%, the time can be obtained by using a larger mass ratio of MTMDs.
amplitude at the first resonant frequency obviously sinks down However, the late response is bigger, and it will take a long
and new peaks appear at the two sides. This is also crucial to time for the frame to ease down completely, which may be not
3400 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

(a) Near the first resonant frequency. (b) Near the second resonant frequency.

Fig. 10. Frequency response of bending strain of the mid-section of member 8–9 about the X -axis.

pertaining to the poles in the integrand of the Fourier integral.

Comprehensive numerical comparisons are made with fine
FEM models, indicating that the proposed method works
very well for free vibration, frequency response, and transient
response analyses.
By comparing the current RMM for structures with MTMDs
with the finite element analysis, we find:
(1) The RMM is uniform in formulations for both structures
with and without MTMDs. On the contrary, extra spring
and damper elements must be incorporated into the FEM
model of structures with MTMDs in general commercial
FEM software such as ANSYS.
(2) The RMM is based on a continuum model of structure,
rather than the approximate discrete model in the FEM.
Therefore, the RMM is more efficient for high-frequency
Fig. 11. Transient response of bending strain of the mid-section of member 8–9 analysis, for which a finer FEM mesh must be used to get
about the X -axis.
precise results, imposing a big burden on the computation.
beneficial to a practical structure due to the associated problem The computational efficiency is especially important for the
of fatigue. optimum design of MTMDs for which the computation will
be repeated many times.
6. Conclusions (3) When the FEM is used for structures with MTMDs
combined with the mode superposition method, the global
In this paper, the RMM is applied and extended to the damping matrix will not be of the Rayleigh type, because
dynamic analysis of structures with MTMDs. The effect of the damping of the main structure is generally not the
TMDs is considered in forming the scattering relations at same as those of MTMDs. This will cause difficulties for
joints, but the unknown displacements of the TMDs are decoupling and solving the dynamic equations.
eliminated properly to make the formulations of the RMM
the same as those for structures without MTMDs. The theory Based on the main findings listed above, we can conclude
of generalized inverse matrices is employed to calculate the that RMM, when properly combined with other techniques
frequency response of structures, which can predict precise (e.g. the generalized inverse matrix, Neumann series expansion,
results even when the excitation frequency is very near to artificial damping), could be an alternative for dynamic analysis
the natural frequency of an undamped structure. The artificial of structures, especially those with complex configuration or in
damping technique, which is widely employed in structural the high-frequency range.
dynamics analysis, is utilized along with the RMM to obtain
relatively long time transient responses of structures without Acknowledgements
damping or structures with small damping (such as structures
with few TMDs). This successfully avoids the singular behavior The work was supported by the National Natural Science
(or nearly singular behavior in the small damping case) Foundation of China (No. 10432030 and No. 10725210),
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3401

E Aik1 eik1 x 0 0 0 0 0
 
 0 ζ2 eik2 x ζ3 eik3 x 0 0 0 
ζ5 eik5 x ζ6 e ik6 x
 
0 0 0 0
A Jf K
 
= 

 0 0 0 G I x ik4 eik4 x 0 0 

 0 0 0 0 −E I y k52 eik5 x 2 ik6 x 
−E I y k6 e
0 −E Iz k22 eik2 x −E Iz k32 eik3 x 0 0 0

Box I.

E Aik1 e−ik1 x 0 0 0 0 0
 
 0 ζ2 e−ik2 x ζ3 e−ik3 x 0 0 0 
ζ5 e−ik5 x ζ6 e−ik6 x 
 
0 0 0 0
D Jf K

= − 

 0 0 0 G I x ik4 e−ik4 x 0 0 

 0 0 0 0 E I y k52 e−ik5 x 2 −ik6 x 
E I y k6 e
0 E Iz k22 e−ik2 x E Iz k32 e−ik3 x 0 0 0

Box II.

eik1 x 0 0 0 0 0
 
 0 (1 + α2 ) eik2 x (1 + α3 ) eik3 x 0 0 0 
(1 + α5 ) eik5 x (1 + α6 ) eik6 x
   0 0 0 0

AuJ K xJK,ω = 


 0
 0 0 eik4 x 0 0 

 0 0 0 0 ik5 eik5 x ik6 eik6 x 
0 ik2 eik2 x ik3 eik3 x 0 0 0

Box III.

e−ik1 x 0 0 0 0 0
 
 0 (1 + α2 ) e−ik2 x (1 + α3 ) e−ik3 x 0 0 0 
(1 + α5 ) e−ik5 x (1 + α6 ) e−ik6 x
   0 0 0 0

DuJ K x ,ω = 
JK
.

 0
 0 0 e−ik4 x 0 0 

 0 0 0 0 −ik5 e−ik5 x −ik6 e−ik6 x 
0 −ik2 e−ik2 x −ik3 e−ik3 x 0 0 0

Box IV.
 q 
the Specialized Research Fund for the Doctoral Program of 2
k2,3

2
= η1 βv + βφz ± η1 βv − βφz + 4c12 βv /Rz2 /2c12
Higher Education (No. 20060335107) and the Program for New
Century Excellent Talents in University (No. NCET-05-05010). (A.3)
α2,3 = −Rz2 (βφz − c12 k2,3
2
)/(κc22 ),
(A.4)
Appendix A. Local phase matrices of displacement and βv = ω2 − iωηv /ρ, βφz = ω2 − iωηφz /ρ
force vectors  q 

2
2
k5,6 = η1 βw + βφy ± η1 βw − βφy + 4c12 βw /R 2y /2c12
See Boxes I–IV. (A.5)
The elements on the third and fifth rows should have
opposite signs when J > K because z J K and z K J are in α5,6 = −R 2y (βφy − c12 k5,6
2
)/(κc22 ),
(A.6)
the same direction. Other undefined parameters appeared above βw = ω2 − iωηw /ρ, βφy = ω2 − iωηφy /ρ
are
ζ j = iκG Ak j α j , ( j = 2, 3, 5, 6) (A.7)
√
η1 = E/κG,
p
i = −1, c1 = E/ρ,
(A.1) where k j ( j = 1, 2, . . . , 6) are wave numbers for various
R y = I y /A, Rz = Iz /A
p p p
c2 = G/ρ, modes; c1 and c2 are wave speeds of the longitudinal and shear
k12 = (ρω2 − iωηu )/E, k42 = (ρω2 − iωηφx )/G (A.2) waves, respectively.
3402 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403

Appendix B. Coordinate transformation matrix References

An arbitrary pair of points P1 and P2 on the principal plane [1] Housner GW, Bergman LA, Caughey TK, Chassiakos AG, Claus RO,
of inertia ox y are needed in order to determine the coordinate Masri SF, et al. Structural control: Past, present, and future. ASCE Journal
transformation matrix between the local and global coordinate of Engineering Mechanics 1997;123:897–971.
systems given by [2] Xu K, Igusa T. Dynamic characteristics of multiple substructures
under closely spaced frequencies. Earthquake Engineering and Structural
λ11 λ12 λ13
 
Dynamics 1992;21:1059–70.
λ 0
 
[3] Jangid RS. Dynamic characteristics of structures with multiple tuned mass
T= , λ = λ21 λ22 λ23  (B.1)
0 λ dampers. Structural Engineering and Mechanics 1995;3:497–509.
λ31 λ32 λ33 [4] Jangid RS, Datta TK. Performance of multiple tuned mass dampers
where for torsionally coupled system. Earthquake Engineering and Structural
Dynamics 1997;26:307–17.
λ11 = (X K − X J ) /L , λ12 = (Y K − Y J ) /L , [5] Jangid RS. Optimum multiple tuned mass dampers for base-excited
(B.2) undamped system. Earthquake Engineering and Structural Dynamics
λ13 = (Z K − Z J ) /L
1999;28:1041–9.
q
[6] Park J, Reed D. Analysis of uniformly and linearly distributed
L = (X K − X J )2 + (Y K − Y J )2 + (Z K − Z J )2 (B.3) mass dampers under harmonic and earthquake excitation. Engineering
Structures 2001;23:802–14.
g1 = X P2 − X P1
/D, g2 = Y P2 − Y P1 /D,



[7] Gu M, Chen SR, Chang CC. Parametric study on multiple tuned mass
g3 = Z P2 − Z P1
/D when J < K dampers for buffeting control of Yangpu Bridge. Journal of Wind
(B.4)
g1 = X P1 − X P2
/D, g2 = Y P1 − Y P2 /D,


Engineering and Industrial Aerodynamics 2001;89:987–1000.
g3 = Z P1 − Z P2 /D when J > K [8] Rana R, Soong TT. Parametric study and simplified design of tuned mass
q dampers. Engineering Structures 1998;20:193–204.

2
2
2 [9] Lei Z, Samir AN. Optimization of the individual stiffness and damping
D= X P2 − X P1 + Y P2 − Y P1 + Z P2 − Z P1 (B.5)
parameters in multiple-tuned-mass-damper systems. Proceedings of SPIE,
A1 = λ12 g3 − λ13 g2 , A2 = λ13 g1 − λ11 g3 , Smart Structures and Materials 2003;5052:217–29.
(B.6) [10] Muhammad NSH, Arfladi YY. Optimum design of absorber for MDOF
A3 = λ11 g2 − λ12 g1 structures. ASCE Journal of Structural Engineering 1998;124(11):
1272–80.
λ31 = A1 /A, λ32 = A2 /A, λ33 = A3 /A,
[11] Joshi AS, Jangid RS. Optimum parameters of multiple tuned mass
q (B.7) dampers for base-excited damped systems. Journal of Sound and
A = A21 + A22 + A23
Vibration 1997;202(5):657–67.
λ21 = (λ13 A2 − λ12 A3 ) /A, [12] Bakre SV, Jangid RS. Optimum parameters of tuned mass damper for
damped main system. Structural Control and Health Monitoring 2007;14:
λ22 = (λ11 A3 − λ13 A1 ) /A, (B.8) 448–70.
λ23 = (λ12 A1 − λ11 A2 ) /A [13] Kwona SD, Park KS. Suppression of bridge flutter using tuned
mass dampers based on robust performance design. Journal of Wind
where X i , Yi , Z i (i = K , J, P1 , P2 ) are the global coordinates Engineering and Industrial Aerodynamics 2004;92:919–34.
of the respective points. By the definition of g1 , g2 and g3 [14] Li CX. Performance of multiple tuned mass dampers for attenuating
given in Eq. (B.4), one can guarantee that z J K and z K J are undesirable oscillations of structures under the ground acceleration.
Earthquake Engineering and Structural Dynamics 2000;29:1405–21.
in the same direction while x J K and x K J as well as y J K and
[15] Wu JN, Chen GD, Lou ML. Seismic effectiveness of tuned mass
y K J are opposite to each other. In the application of the above dampers considering soil-structure interaction. Earthquake Engineering
formulation, the line P1 P2 shall be selected not to parallel the and Structural Dynamics 1999;28:1219–33.
member JK. [16] Li CX, Liu YX, Wang ZM. Active multiple tuned mass dampers: A new
control strategy. ASCE Journal of Structural Engineering 2003;129(7):
Appendix C. Generalized inverse matrix 972–7.
[17] Li CX, Liu YX. Active multiple tuned mass dampers for structures
The Moore–Penrose generalized inverse matrix G+ of any under the ground acceleration. Earthquake Engineering and Structural
matrix G ∈ Cm×n uniquely exists, satisfying Dynamics 2002;31:1041–52.
[18] Li CX. Multiple active–passive tuned mass dampers for structures
GG+ G = G, G+ GG+ = G+ , under the ground acceleration. Earthquake Engineering and Structural
H
H (C.1) Dynamics 2003;32:949–64.
GG+ = GG+ , G+ G = G+ G.


[19] Lin JH, Zhang WS, Sun DK, He Q, Lark RJ, Williams FW. Precise and
efficient computation of complex structures with TMD devices. Journal of
According to the singular value decomposition (SVD) theory,
Sound and Vibration 1999;223(5):693–701.
there exist unitary matrices U and V such that [20] Chen YH, Chen DS. Timoshenko beam with tuned mass dampers
to moving Loads. ASCE Journal of Bridge Engineering 2004;17:
   −1 
F 0 F 0
G=U V ,
H +
G =V UH 167–77.
0 0 m×n 0 0 n×m [21] Lin CC, Wang JF, Chen BL. Train-induced vibration control of high-
(C.2) speed railway bridges equipped with multiple tuned mass dampers. ASCE
√ √ √ Journal of Bridge Engineering 2005;10(4):398–414.
where F = diag[ λ1 , λ2 , . . . , λr ], with λi (i = [22] Li JZ, Su MB, Fan LC. Vibration control of railway bridges under high-
1, 2, . . . , r ) being the positive eigenvalues of G H G and r the speed trains using multiple tuned mass dampers. ASCE Journal of Bridge
rank of matrix G. Engineering 2005;10(3):312–20.
 Y.Q. Guo, W.Q. Chen / Engineering Structures 29 (2007) 3390–3403 3403

[23] Howard SM, Pao YH. Analysis and experiments on stress waves in planar [28] Chopra AK. Dynamics of structures. Upper Saddle River (NJ): Prentice-
trusses. ASCE Journal of Engineering Mechanics 1998;124:884–91. Hall; 2001.
[24] Pao YH, Keh DC, Howard SM. Dynamic response and wave propagation [29] Horn RA, Johnson CR. Matrix analysis. New York: Cambridge University
in plane trusses and frames. AIAA Journal 1999;37:594–603. Press; 1985.
[25] Chen JF, Pao YH. Effects of causality and joint conditions on method of [30] Humar JL, Xia H. Dynamic response analysis in frequency domain.
reverberation-ray matrix. AIAA Journal 2003;41:1138–42. Earthquake Engineering and Structural Dynamics 1993;22:1–12.
[26] Pao YH, Sun G. Dynamic bending strains in planar trusses with pinned or [31] Humar JL, Bagchi A, Xia H. Frequency domain analysis of soil-structure
rigid joints. ASCE Journal of Engineering Mechanics 2003;129:324–32. interaction. Computers & Structures 1998;66:337–51.
[27] Pestel EC, Leckie FA. Matrix methods in elasto mechanics. New York [32] Kausel E, Roesset JM. Frequency domain analysis of undamped systems.
(NY): McGraw-Hill; 1963. ASCE Journal of Engineering Mechanics 1992;118(4):721–34.