Engineering Structures 159 (2018) 89–98

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

A novel bat algorithm based optimum tuning of mass dampers for improving T
the seismic safety of structures
⁎
Gebrail Bekdaşa, Sinan Melih Nigdelia, , Xin-She Yangb
a
Department of Civil Engineering, Istanbul University, 34320 Avcılar, Istanbul, Turkey
b
Design Engineering and Mathematics, Middlesex University, The Burroughs, London, UK

A R T I C L E I N F O A B S T R A C T

Keywords: Metaheuristic algorithms are effective for optimization with diverse applications in engineering. The optimum
Tuned Mass Damper tuning of tuned mass dampers is very important for seismic structures excited by random vibrations, and op-
Optimization timization techniques have been used to obtain the best performance for optimally tuned mass dampers. In this
Metaheuristic algorithms study, a novel optimization approach employing the bat algorithm with several modifications for the tuned mass
Bat algorithm
damper optimization problem is presented. In the proposed method, the design variables such as the mass,
Harmony search algorithm
period and damping ratio of tuned mass damper are optimized and different earthquake records are considered
during the optimization process. The method is then applied to a ten-story civil structure and the results are then
compared with the analytical methods and other methods such as genetic algorithms, particle swarm optimi-
zation, and harmony search. The comparison shows that the proposed method is more effective than other
compared methods. Additionally, the robustness of the optimum results was evaluated. The proposed approach
for optimizating tuned mass dampers via the bat algorithm is a feasible and efficient approach.

1. Introduction system, without inherent damping and excited by harmonic excitations.

The representation of simple systems like beams and plates as a single
The vibration absorber device invented by Frahm [1] was modified of degree of freedom systems may be a good assumption and the op-
with damping elements in order to damp random vibrations, and the timum parameters of TMDs were presented by Warburton and Ayorinde
tuned mass dampers (TMDs) were generated [2]. Thus, all mechanical [4] for light damping in the simple systems. Also, Warburton [5] pro-
systems subject to undesired random vibrations can be protected. Since posed the expressions for the same parameters for harmonic and white
civil structures are big mechanical systems that are also exposed to noise excitations. An analytical formula is not possible for main struc-
undesired excitations such as strong winds, earthquakes and moving tures with inherent damping. For that reason, the optimum TMDs can
traffic, tuned mass dampers have been installed in many important be designed by using numerical iterations [6,7]. Sadek et al. obtained
buildings. In addition to prevent structural failure, there are many the optimum tuning expressions by using curve fitting to numerically
reasons to use TMDs in structures, including the protection of sensible searched values and modifying the expression for systems with multiple
equipment and even preventing motion sickness of residents. Examples degrees of freedom (MDOF) [8]. Chang [9] proposed several expres-
include the Taipei 101 building in Taipei, the Television Tower in sions for different types of TMDs. Rüdinger [10] investigated a TMD
Berlin, John Hancock Tower in Boston, Lax Theme Building in Los with a nonlinear power law viscous damper excited by white noise.
Angeles, Burj Al-Arab in Dubai and the Millennium Bridge in London. Marano et al. [11] optimized TMDs for seismic effects by using a con-
For the best performance, TMDs must be optimally tuned and the op- strained reliability-based optimization minimizing the maximum of the
timization formulation must consider the characteristics of structures dimensionless peak of displacement of the protected system with re-
and excitations. spect to the unprotected one under different input characteristics.
For the tuning of TMDs, several approaches have been proposed in Recently, several studies about mass dampers have been developed.
the literature. For example, Den Hartog [3] developed some optimum Mensah and Dueñas-Osorio proposed tuned liquid column dampers
tuning formulas for frequency and damping ratios of TMDs by the (TLCDs) in order to improve the reliability of wind turbines [12]. The
change of the ratio of the masses of TMDs to the structure under con- optimum tuning formulation for TLCDs were developed by Di Matteo
sideration. These equations were derived for a single degree of freedom et al. [13] for random vibrations. Salvi and Rizzi proposed a numerical

⁎
Corresponding author.
E-mail address: [email protected] (S.M. Nigdeli).

https://doi.org/10.1016/j.engstruct.2017.12.037
Received 15 April 2017; Received in revised form 17 December 2017; Accepted 19 December 2017
0141-0296/ © 2017 Elsevier Ltd. All rights reserved.
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Minimax optimization for optimum tuning of mass dampers in order to

provide a good performance for frames under earthquake excitations xd
[14]. Lu et al. proposed a new eddy-current tuned mass damper in order
to improve the performance of tall structures [15]. Lavan developed a
kd
TMD md
multi-objective optimum design procedure in order to reduce structural
xN cd
responses, TMD mass and stroke [16]. Pendulum type TMDs were also rigid mass
proposed for seismic performance of power plants [17]. Lu et al. con-
ducted experimental studies for particle TMDs for wind-induced vi-
mN
brations [18]. cN
The nature inspired metaheuristic algorithms have a wide applica- kN
tion area [19–22] and these algorithms have also been attempted for
the optimization of TMDs. The employed metaheuristic methods in the
literature are genetic algorithm (GA) [23–27], bionic algorithm [28],
particle swarm optimization (PSO) [29,30] and harmony search (HS) xi rigid mass
[31–34]. Although the optimization of TMDs have been discussed in
several studies for seismic control of structures, it is still an important mi
area that needs to be investigated since these methods use several cri- ci
teria and procedures. As it is well known that earthquake excitations
are random vibrations which contain different pulses like motions with
ki
different frequencies, an optimally tuned TMD should always show the
best performance for different earthquake excitations. For that reason,
designers must find the best TMD design which is still robust when a
x1
different excitation is applied to structure. Because of the unique rigid mass
characteristics of the bat algorithm defined in step 4 of Section 2, it is a
suitable metaheuristic algorithm for analysing and optimizing TMD
m1
parameters in an iterative manner. One of the main advantages of the c1
bat algorithm is its ability of finding the global optimality with quick k1
convergence [35,36].
In order to provide sufficient damping of seismic vibrations, the
characteristics of a TMD must be tuned so as to achieve the required,
special properties of passively controlled system. The tuning of TMDs is
a process in which the optimum parameters are found, often in an ..
iterative manner. The parameters of a TMD are related with its fre- xg
quency, mass and damping coefficient. In Fig. 1, a multi-story structure
Fig. 1. Physical model of a structure with TMDs.
with TMDs is shown by employing a degree of freedom in the trans-
actional direction for each story and TMD.
In several existing studies, simple expressions were proposed for the TMD optimization problem. A novel methodology is developed in order
optimum frequency ratio (fopt) and damping ratio (ξd,opt). These ex- to find the optimal values of the design variables such as the mass of
pressions depended on a constant mass ratio (μ) and only a single TMD (md), period of TMD (Td) and damping ratio of TMD (ξd) for multi
(critical) vibration mode was considered. Several formulas of the op- story seismic structures. Different from the other optimization algo-
timum TMD frequency and damping ratio are given in Table 1. rithms using equations for the objective function, BA is combined with
In the close form expressions, the damping ratio (ξ) of the main the numerical analyses process of seismic structures and time domain
structure is considered in the approach by Sadek et al. [8] and Leung histories of the structures are considered as optimization objective re-
and Zhang [30]. By finding fopt and ξd,opt, the optimum frequency of the sults. Since the optimization problem does not have explicit, analytical
TMD (wd,opt) and the optimum damping coefficient of TMD (cd,opt) are solutions, the nonlinearity of this problem necessitates a novel ap-
found for a structure with the frequency described as ws. Since the close proach by modifying the use of the bat algorithm so as to be suitable for
form expressions were derived for a single degree of freedom (SDOF) a dynamic analyses problem and the tuning of TMD parameters for
system, a multiple degrees of freedom (MDOF) system must be idealized structures subject to excitations with wide variety of frequencies. In
and be related to a SDOF system. In the idealization of the structure, the addition, we will compare the results obtained by our proposed ap-
modal mass for the first critical frequency is taken as the mass of the proach with those obtained by other methods. Therefore, this paper is
structure in calculating the mass ratio. Also, according to the proposal organized as follows. Section 2 provides the detailed formulation of the
of Sadek et al. [8], the frequency ratio for MDOF systems is nearly equal TMD problem, and discusses the optimization methods. Then, Section 3
to the ratio for SDOF systems with a modified mass ratio as μΦ. Also, provides some numerical examples, followed by discussions and con-
the optimum damping ratio for SDOF systems is multiplied by Φ for the clusions in Section 4.
modification of MDOF systems. Here, Φ is the amplitude of the first
mode shape at a location where the TMD is located. This amplitude is 2. Design and optimization methodology
calculated for the unit modal participation factor (Γi).
The use of close form expressions for MDOF systems is essentially The formulation of the TMD problem will be described along with
based on some assumptions and idealizations that may not be exactly the description of the optimization procedure. Three design variables to
valid in practice. In contrast, by using proper metaheuristic algorithms be optimized are the mass of TMD (md), period of TMD (Td) and
such as the bat algorithm, it is possible to find the optimum parameters damping ratio of TMD (ξd). The optimization algorithm to be used is
for multi-story structures subject to random excitations such as earth- mainly the bat algorithm, and the comparison will be made with other
quake-induced ground motions. Therefore, a methodology employing algorithms and close form expressions.
the Bat Algorithm is proposed for solving the present TMD optimization In the methodology, the bat algorithm is combined with the dy-
problem. namic analysis process of seismic structures to generate an effective
In this paper, the recent bat algorithm (BA) is modified to solve the novel optimization approach for tuning of TMDs. The proposed

90
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Table 1
The frequency and damping ratio expressions of the compared methods.

Method wd,opt cd,opt Governing excitation

fopt = ξd,opt =
ws 2md wd,opt

Den Hartog [3] 1 3μ Harmonic excitations

1+μ 8(1 + μ)
Warburton [5] 1 − (μ / 2) μ (1 − μ / 4) White noise excitations
1+μ 4(1 + μ)(1 − μ / 2)
Sadek et al. [8] 1 ⎡ μ ξ μ Frequency domain analyses
1−ξ ⎤ +
1+μ 1+μ 1+μ 1+μ
⎣ ⎦
Leung & Zhang (PSO approach) [30] 1 − (μ / 2) μ (1 − μ / 4) White noise excitation
1+μ 4(1 + μ)(1 − μ / 2)
+ (−4.9453 + 20.2319 μ −37.9419μ) μ ξ − 5.3024ξ 2μ
+ (−4.8287 + 25.0000 μ ) μ ξ 2

methodology can be summarized as the following six steps: Step 3: Bats in nature use echolocation to sense distance and to
search for prey, and these bats also use frequency tuning by varying the
Step 1: In the first step, the properties of the main structure, ranges frequency, loudness and pulse emission rates when homing for preys.
of design variables, earthquake excitations and parameters of the Such characteristics are used in the bat algorithm where the location of
bat algorithm are defined. The bat algorithm (BA) developed by a bat corresponds to the position vector. In our problem, the position
Yang [35] was formulated by idealizing the echolocation behaviour vectors used in the bat algorithm (di) from i = 1 to n are generated by
of bats. The parameters and bounds used in BA are the bat popu- randomization as design variables. The designed variables are assigned
lation (n), the limits of pulse frequency (fmin and fmax), the pulse with the values within the defined ranges/limits. Also, the objective
emission rate (ri) and the loudness (Ai). The optimization process function for all position vectors is calculated and stored. The objective
must be conducted for various earthquake excitations as we do in of the optimization problem is to minimize the ratio of the maximum
this study. Otherwise, the optimum parameters may be a local op- first-story displacements (x1) of the structure with and without TMD
timum for all the earthquake excitations. (RD). The objective function cannot be written with a formula respect
Step 2: In this step, the dynamic analyses of the structure without to the design variables as done in the classical optimization approaches
TMD are done for all earthquake excitations. These analyses results employing heuristic algorithms. For that reason, the BA based metho-
are used in comparison of the performance of the TMD. The equa- dology is combined with the dynamic analyses procedure described in
tions of motion of the structure are modelled via the Matlab Step 2 and the objective function is provided according to the analyses
Simulink [37]. In the matrix form, the main governing equation of results of structures with and without TMD. In the analyses of structure
the structure can be written as with TMD, the candidate solutions generated by the BA are used. The
analyses of structure without TMD is done in Step 2, but the analyses of
Mx¨ (t ) + Cx ̇ (t ) + Kx (t ) = −M{1} x¨g (t ) (1)
structure with TMD must be done for all candidate solutions. A position
where the mass, damping and stiffness matrices, respectively, are re- vector in the bat algorithm and objective function are, respectively,
presented by M, C and K in Eqs. (2)–(4) for an N story structure. The given by Eqs. (6) and (7).
dots on x(t) represent the derivatives of the function in time. In addi- di = {m di; Tdi;ξdi} for i = 1 to n (6)
tion, the three matrices can be written as
M= diag[m1 m2…mN ], (2) max(first story displecement of structure with TMD)
fi (d) = RD = for i
max(first story displecement of structure without TMD)
⎡ (c1 + c2) − c2 ⎤ = 1 to n (7)
⎢ − c2 (c2 + c3) − c3 ⎥
⎢ . . ⎥ The objective function is evaluated by conducting dynamic analyses
C= ⎢ . . . ⎥, as explained in Step 2, but a new degree of freedom is added for the
⎢ . cN −1 (cN −1 + cN ) − cN⎥
⎢ − cN cN ⎥ TMD, so M, C, K and x(t) are updated as given by the following equa-
⎣ ⎦ (3) tions [Eqs. (8)–(11)].
M= diag[m1 m2…mN md ] (8)
⎡ (k1 + k2) − k2 ⎤
⎢ − k2 (k2 + k3) − k3 ⎥
K= ⎢ . . ⎥.
. . . ⎡ (c1 + c2) − c2 ⎤
⎢ ⎥ ⎢ − c2 ⎥
⎢ . kN − 1 (kN − 1 + kN ) − kN ⎥ (c2 + c3) − c3
⎢ . . ⎥
⎢ − kN kN ⎥ (4) C= ⎢ . . . ⎥
⎣ ⎦ . . .
⎢ − cN (cN + cd ) − cd ⎥
The response vector for the structure is defined as ⎢ ⎥
⎢ − cd cd ⎥
⎣ ⎦ (9)
x (t ) = [x1 x2…xN ]T (5)

In addition, ẍ denotes the ground acceleration (earthquake data) ⎡ (k1 + k2) − k2 ⎤

and xi defines the displacements of ith story of the structure. In the ⎢ − k2 (k2 + k3) − k3 ⎥
⎢ . . ⎥
above equations, mi, ci and ki values are mass, damping coefficient and K= ⎢ . . . ⎥
. . .
stiffness coefficient of ith story of main structure, respectively. Here, N ⎢ − kN (kN + kd ) − kd ⎥
⎢ ⎥
is the number of stories. ⎢ − kd kd ⎥
In the dynamic analyses, the conventional Runge-Kutta method ⎣ ⎦ (10)
with a time step of 0.001 s is used. In addition to time domain analyses,
x (t ) = [x1 x2…xN x d]T (11)
frequency domain analyses are also conducted as a second criterion for
optimization. where xd is the displacement of the TMD, while md, cd and kd are the

91
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

mass, damping coefficient and stiffness coefficient of TMD, respectively. if (random number > ri)
Here, kd and cd are defined as the design variables as given in Eqs. (12)
and (13), respectively. Select a position vector
Generate a local solution around the selected position vector
md
k d = wd2 md = 4π 2 end if
Td2 (12)
if (random number < Ai & f(di) < f(d∗))
Accept the updated position vectors
kd ξ md
cd = 2ξd md = 4π d Increase ri and reduce Ai
md Td (13) end if
Also, the analyses in the frequency domain are conducted and the Step 6: Different from the classical bat algorithm, iterations are not
maximum first story acceleration transfer function is recorded.
conducted for a constant number of iterations. There are two stop-
ping criteria to finalize the iterative process in the novel metho-
Step 4: In this step, iterative analyses are started. The position vec-
dology. A similar process is conducted with harmony search ap-
tors in the bat algorithm are updated, according to the equations of
proach given in [33].
the bat algorithm to be described below. In the BA, new solutions
are found by adjusting frequency (fi) and updating velocity (vi). The
The value of objective function defined by RD (Eq. (7)) must be
procedure is formulized in Eqs. (14)–(16) where β is a random
smaller than a user defined value for the best solution. The selected
number between 0 and 1.
value is increased after several iterations (or a fixed number of itera-
fi = fmin + (fmax−fmin)β (14) tions) if a solution cannot be found, because the reduction of the dis-
placement of the structure may not be physically possible for the de-
v ti = v ti−1 + (d ti −d∗)fi (15) fined ranges of the design variables, or an applicable solution in
application cannot be found. Also, the value of maximum acceleration
d ti = d ti−1 + v ti (16) transfer function must be smaller than the value of the structure
without the TMD for the structure with a TMD. The transfer function
Here, d∗ is the set of design variables, corresponding to the minimum
value in the frequency domain is independent of earthquake excita-
RD value. Eqs. (14)–(16) are applied to the whole population of bats
tions. By limiting the transfer function value with the value of the un-
(i = 1 to n), and RD values are also updated by conducting dynamic
controlled structure, a feasible solution for different excitations can be
analyses. If the ranges of design variables are exceeded, the position
ensured. Step 4 and Step 5 are repeated until the stopping criteria are
vector (di) is limited with the boundaries of the ranges. In that case, the
satisfied.
limits of the ranges are chosen as new positions.
The methodology is summarized in the flowchart given as Fig. 2.
By using passive control systems like TMDs in seismic structures, the
responses can be only reduced by changing the characteristics of the
modal frequencies. Since the earthquake excitations contain random 3. Case studies
frequencies, the optimum results of TMD parameters can be near to the
existing best solutions. The optimum solution may be too close or a bit 3.1. Ten story shear building
far from the current best values (d∗). Additionally, this solution may be
between the best and a random solution, because the response of the A ten-story shear building was investigated for an optimum TMD on
structure is complex when the inherent damping is considered. For the top of the structure. The properties of the structure for the present
these reasons, BA generates new variable with changing frequency as study [24] are the same for all stories. The mass, stiffness coefficient
seen in Eq. (14) and updates all solutions with respect to the best one as and damping coefficient of a story of the structure are 360 t, 650 MN/m
seen in Eq. (15). and 6.2 MNs/m, respectively. The critical period of the structure (T) is
1 s.
Step 5: In this step, the updated position vectors generated in step 4 The optimum TMD parameters were found by using the proposed
are accepted or not according to the criterion test of the pulse rate methodology via the bat algorithm. The performance of the TMD is
and loudness. If the pulse rate is smaller than a random number in compared with the results of the methodology given in [30] for the
[0,1], local solutions around the best position vector are generated. same structure and excitations. The compared methodology employs
In that case, the algorithm may generate new variables exceeding the HS algorithm developed by Geem et al. [38]. Also, the optimum
the range limits. These values may be also optimum ones, but these parameters are compared with the methods given in Table 1.
values cannot be accepted. The existing of the variables ranges may For dynamic analyses, the equations of motions are directly ana-
be constraints related to economic and physical reasons. The values lysed without separating the vibration modes. The design variables such
of design variables are limited by the limits of the ranges if the as md, Td and ξd were searched for the ranges given in Table 2.
randomly assigned values are out of the ranges. The earthquake excitations used in the optimization of design
variables are given in Table 3 including peak ground acceleration
If a random number is smaller than the loudness and the RD value (PGA), peak ground velocity (PGV), duration, magnitude and the clo-
for the pervious design variables are bigger than the best solution, then sest distance to the fault (R). Generally, the chosen earthquakes are
new positon vectors generated in Step 4 are accepted. In addition, the near-fault records from the different regions of the world in order to
value of pulse rate (ri) is increased during iterations, while the value of find a global optimum TMD values. These earthquake records were
loudness (Ai) is reduced, according to Eqs. (17) and (18). Here, α and γ downloaded from the website of Pacific Earthquake Engineering Re-
are constant values, which are taken as 0.9 in this study. search Center (PEER) [39]. In the methodology, all excitations are used
in the analyses, but the results of the most critical excitation is taken
Ait + 1 = αAit (17)
into consideration. The critical excitation may chance according to
possible candidate values of TMD parameters. Thus, an effective TMD
rit + 1 = ri0 [1−exp(−γt )]. (18)
for all excitations are found.
The procedure of this step is summarized in the pseudo as below. In the bat algorithm, different initial pulse rate (ri0) and initial
loudness (Ai0) were tested and RD values are presented in Table 4.
According to the results, the bat population (n), the minimum

92
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Start

Define the structure properties, design variable ranges, earthquake

excitations, BA parameters

Analyse the structure without TMD

Generate the initial position vector and perform dynamic analyses for
the candidate solutions

Are the stopping Yes

criteria provided? Stop

No

Generate updated solution vectors according to BA formulations

Otherwise Check the Random number>ri Generate local

Keep the existing
acceptance solutions
solutions
criteria

Random number<Ai
and f(di)<f(d*)
Accept the updated solution and modify r i and Ai

Fig. 2. The flowchart of the methodology.

Table 2 frequency (fmin), the maximum frequency (fmax), the initial pulse rate
Ranges of design variables. (ri0) and initial loudness (Ai0) are taken as 5, 0, 1, 0.5 and 1, respec-
tively. In the HS algorithm, the harmony memory size (HMS), the
Minimum Maximum
harmony memory considering rate (HMCR) and the pitch adjusting rate
md (t) N
0.01 ∑i = 1 mi = 36t
N
0.1 ∑i = 1 mi = 360t (PAR) were taken as 5, 0.5 and 0.2, respectively.
Td (s) 0.8 T = 0.8 s 1.2 T = 1.2 s The optimum results are summarized in Table 5, including the RD
ξd 0.05 0.30 value for the most critical excitation. The optimum values were found
by starting the optimization process for 30 times in order to check the
robustness of the optimization method and the results contain the best

Table 3
Earthquake records used in the analyses and optimization.

Earthquake Station Component PGA (g) PGV (cm/s2) Duration (s) Magnitude (M) R (km)

Kobe KJMA KJM000 0.821 81.3 48 6.9 0.96

Cape Mendocino Petrolia PET090 0.662 89.7 36 7.01 8.18
Erzincan Erzincan ERZ-NS 0.515 83.9 21 6.7 2.00
Northridge Sylmar SYL360 0.843 129.1 40 6.69 5.3
Northridge Rinaldi RRS228 0.838 166.1 20 6.69 6.5
Loma Prieta LGPC LGP000 0.563 94.8 25 6.93 6.1

93
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Table 4
RD values for different initial pulse rate (ri0) and initial loudness (Ai0).

A0/r0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

0.1 0.8157 0.8153 0.8129 0.8100 0.8090 0.8127 0.8141 0.8157 0.8157 0.8158
0.2 0.8146 0.8137 0.8116 0.8096 0.8076 0.8108 0.8135 0.8140 0.8154 0.8157
0.3 0.8142 0.8112 0.8127 0.8083 0.8075 0.8120 0.8099 0.8136 0.8150 0.8157
0.4 0.8138 0.8123 0.8110 0.8081 0.8075 0.8116 0.8103 0.8145 0.8131 0.8157
0.5 0.8137 0.8105 0.8120 0.8080 0.8073 0.8089 0.8129 0.8111 0.8144 0.8156
0.6 0.8136 0.8114 0.8077 0.8099 0.8066 0.8087 0.8108 0.8127 0.8141 0.8154
0.7 0.8132 0.8110 0.8095 0.8077 0.8065 0.8085 0.8126 0.8106 0.8139 0.8154
0.8 0.8130 0.8095 0.8077 0.8087 0.8064 0.8078 0.8092 0.8118 0.8136 0.8153
0.9 0.8121 0.8093 0.8086 0.8069 0.8064 0.8074 0.8089 0.8114 0.8132 0.8143
1.0 0.8118 0.8087 0.8078 0.8066 0.8062 0.8071 0.8086 0.8096 0.8126 0.8140

Table 5 and average of RD of these runs. For all runs of the methodology, the
Optimum results. method must be effective. The standard derivatives, number of total
analyses to reach the optimum result are also presented for two com-
BA HS Den Hartog Warburton Sadek et al. PSO
pared methods. The optimum values for closed form expressions are
md (t) 360.00 357.44 360.00 360.00 360.00 360.00 calculated for the modal mass with the consideration of the multi-de-
Td (s) 0.9733 0.9714 1.1369 1.1818 0.8704 0.7976 grees of freedom modifications. The modal mass, mass ratio, inherent
ξd 0.0974 0.1017 0.1927 0.1614 0.3347 0.1606 damping ratio of the structure and Φ are calculated as 1900.614 t,
Best RD 0.8061 0.8106 0.8403 0.8361 0.9115 0.9077
18.9%, 3% and 0.79, respectively.
Average RD 0.8097 0.8162 – – – –
Standard 0.0019 0.0036 – – – – As seen clearly from the table, the proposed methodology by BA is
Deviati- more effective on the reduction of the RD value for the most critical
on earthquake (Nortridge-Sylmar excitation) than the HS-based or PSO-
Analyses for 1576 1783 – – – –
based approaches and close form expressions. BA is also effective on
op-
timum computation time and standard derivative values. T-test is applied for
30 independent runs. In order to test average value (µ) of proposed
method is less than HS approach (0.81617), the flowing hypothesis can

Cape Mendocino Kobe Fig. 3. Time history plots for optimization earthquake
0.05 0.08 excitations.

0.025 0.04
x1 (m)

x1 (m)

0 0

-0.025 -0.04

-0.05 -0.08
0 10 20 30 40 0 10 20 30 40
Time (s) Time (s)
Erzincan Northridge-Rinaldi
0.05 0.1

0.025 0.05
x1 (m)

x1 (m)

0 0

-0.025 -0.05

-0.05 -0.1
0 10 20 30 0 10 20 30
Time (s) Time (s)
Northridge-Sylmar Loma Prieta
0.06 0.06

0.03 0.03
x1 (m)

x1 (m)

0 0

-0.03 -0.03

-0.06 -0.06
0 10 20 30 0 10 20 30
Time (s) without TMD with TMD
Time (s)

94
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Fig. 4. Maximum story displacements for optimization earthquakes.

Table 6
Earthquake records for benchmark.

Earthquake Station Component PGA (g) PGV (cm/s2) Duration (s) Magnitude (M) R (km)

Landers Lucerne LCN000 0.785 31.9 48 6.9 0.96

Kocaeli Sakarya SKR090 0.376 79.5 60 7.51 3.12
San Fernando Pacoima Dam PUL164 1.17 113.5 41.6 1.81 6.61
Imperial Valley El-Centro I-ELC180 0.313 29.8 40 6.95 6.09
Tabas Tabas TAB-TR 0.852 121.4 32.8 7.35 2.05
Düzce Bolu BOL090 0.822 62.1 55.9 7.14 12.04

95
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Fig. 5. Maximum story displacements for benchmark earthquakes.

Table 7
RD values for stiffness change by ± 20%.

Stiffness 80% 85% 90% 95% 100% 105% 110% 115% 120%

RD 0.9735 0.9334 0.8895 0.8427 0.8061 0.8230 0.8245 0.8156 0.8339

Table 8
The optimum results (maximum 108.0 t mass).

Den Hartog Warburton Sadek et al. PSO GA BA HS

md (t) 108.00 108.00 108.00 108.00 108.00 108.00 104.40

Td (s) 1.0337 1.0455 0.9529 0.9213 1.0663 1.1795 1.1860
ξd 0.1120 0.0921 0.2055 0.0919 0.1190 0.0500 0.0620
RD 0.9961 0.9979 0.9903 1.0059 0.9878 0.9463 0.9486

96
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

Table 9 order to test the robustness of the optimum design of the TMD, more
Properties of the 40 story structure [40]. earthquakes are used, and the information about the benchmark re-
cords are given in Table 6.
mi (t) 980
k1-k40 (MN/m) 2130–998 The maximum story displacements are illustrated in Fig. 5 for these
c1-c40 (MNs/m) 42.6–20 benchmark earthquakes. The optimum TMD is also effective for the
benchmark earthquakes. The reduction percentages of the first story
displacements are between 20.7 and 43.3.
be used: The properties of the main structure may differ because of un-
H0 : μ = 0.81617 (19) certainties. For that reason, the robustness of optimum TMD was
checked by changing the stiffness of the structure within certain per-
H1: μ < 0.81617 (20) centages. In Table 7, the RD values are given for ± 20% change of the
x −μ 0.8097−0.8162 stiffness of the main structure.
tstat = = = −18.73 The optimum TMD is robust as it can be seen from the results given
s/ n 0.0019/ 30 (21)
in Table 5. The effectiveness of TMD is only affected with 2.8% by the
where x is the average value of the proposed method, s is the standard increase of the stiffness. Also, the performance lost is not too much until
deviation and n is the number of observation. For α = 5% significance the 10% decrease of the stiffness of the structure.
level and degree of freedom = 29, t critical value (tcr) is −1.699. Since The same case study was previously investigated by Hadi et al. [20]
tstat < tcr, we reject the H0 for 5% significance level. Thus, the average employing GA. In their GA approach, only the optimum stiffness and
values of the proposed method is better than HS approach with 95% damping coefficient are found for a constant mass. The GA parameters
confidence level. were taken as 30, 200, 0.45 and 0.01 for the population size, number of
The time history plots for the first story displacement under opti- generation, probability of crossover and probability of mutation, re-
mization excitations are shown in Fig. 3 for the structure without the spectively. The constant mass of a TMD (108.0 t) was taken as 3% of the
TMD (wo TMD) and with the TMD optimized by using the proposed total mass of the structure. For that reason, a new case study was in-
approach. vestigated by limiting the mass with 108.0 t in order to compare with
The optimum TMD is effective on reducing the maximum dis- the GA-based approach. The optimum results and RD values are given
placement of the structure for all earthquakes. Also, the optimum TMD in Table 8.
provides the steady-state response for the structure as seen in time The proposed methodology is also more effective than other
history plots. In Fig. 4, the maximum displacements of all stories are methods for cases with smaller maximum mass limits than the first case
given all earthquakes used in the optimization process. study. But, a TMD with 108.0 t mass is not sufficient in order to obtain a
The first story displacements are reduced by 30.5%, 42.8%, 26%, good reduction under Loma Prieta excitation, representing near-fault
25.4%, 19.4% and 19.4% for Cape Mendocino, Kobe, Erzincan, characteristics such as impulsive motions.
Northridge- Rinaldi, Northridge- Sylmar and Loma Prieta excitations,
respectively. The optimum TMD must be effective on different excita-
tions since the ground motion are not negligible before the earthquake. 3.2. Forth story shear building
Furthermore, the optimum results were also tested for benchmark
earthquakes, which are not considered in the optimization process. In The BA based method is also tested for a 40 story structure. The

wo TMD BA Fig. 6. Maximum story displacements of 40 story

1.8 1.8 structure for optimization earthquakes.

Cape Mendocino
Kobe
1.6 1.6 Erzincan
Northridge-Rinaldi
Northridge-Sylmar
1.4 1.4 Loma Prieta

1.2 1.2
max(x ) (m)

max(x ) (m)

1 1
i

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
5 10 15 20 25 30 35 40 5 10 15 20 25 30 35 40
story story

97
G. Bekdaş et al. Engineering Structures 159 (2018) 89–98

properties of the structure are given in Table 9. The mass of all stories [10] Rüdinger F. Optimal vibration absorber with nonlinear viscous power law damping
(mi for i = 1–40) are the same, but the stiffness coefficient of a story (ki) and white noise excitation. J Eng Mech 2006;132:46–53.
[11] Marano GC, Greco R, Trentadue F, Chiaia B. Constrained reliability-based optimi-
linearly decreases from first to top story. The damping coefficients are zation of linear tuned mass dampers for seismic control. Int J Solids Struct
0.02 times of the stiffness coefficients. The optimum values of md, Td 2007;44:7370–88.
and ζd are found as 3905.45 t, 3.8003 s and 0.2608, respectively. The [12] Mensah Akwasi F, Dueñas-Osorio Leonardo. Improved reliability of wind turbine
towers with tuned liquid column dampers (TLCDs). Struct Saf 2014;47:78–86.
effectiveness of the optimum results can be clearly seen from the [13] Di Matteo A, Iacono FL, Navarra G, Pirrotta A. Optimal tuning of tuned liquid
maximum story displacements given as Fig. 6. column damper systems in random vibration by means of an approximate for-
mulation. Meccanica 2015;50(3):795–808.
[14] Salvi Jonathan, Rizzi Egidio. Optimum tuning of Tuned Mass Dampers for frame
4. Discussions and conclusions structures under earthquake excitation. Struct Cont Health Monitor
2015;22(4):707–25.
In the practical application of the TMDs, there are several factors. [15] Lu X, Zhang Q, Weng D, Zhou Z, Wang S, Mahin SA, Ding S, Qian F. Improving
performance of a super tall building using a new eddy-current tuned mass damper.
The most important factor is the additional weight of TMD for the
Struct Control Health Monitor 2017; 24(3).
structure. In the optimization process, the vertical response of the [16] Lavan Oren. Multi-objective optimal design of tuned mass dampers. Struct Control
structure is investigated. The axial loading capacity of the structural Health Monitor 2017.
members must not exceed. This factor is considered by the maximum [17] Shu Z, Li S, Zhang J, He M. Optimum seismic design of a power plant building with
pendulum tuned mass damper system by its heavy suspended buckets. Eng Struct
value of the range of the mass. Additionally, the stroke capacity of 2017; 136: 114–132.
TMDs can be considered in practical applications. Also, the frequency [18] Lu Zheng, Wang Dianchao, Zhou Ying. Experimental parametric study on wind-
content of the excitation is not exactly known for the earthquake re- induced vibration control of particle tuned mass damper on a benchmark high-rise
building. Struct Design Tall Special Build; 2017.
cords. For that reason, an optimum design in needed for several ex- [19] Tsai PW, Zhang J, Zhang S, Istanda V, Liao LC, Pan JS. Improving swarm in-
citation instead of a single one. telligence accuracy with cosine functions for evolved bat algorithm. J Inform Hid
A recently developed metaheuristic bat algorithm has been modified Multimed Signal Process 2015;6(6):1194–202.
[20] Chen YH, Huang HC. Coevolutionary genetic watermarking for owner identifica-
for the optimum tuning problem of the tuned mass dampers used in tion. Neural Comput Appl 2015;26(2):291–8.
civil structures in order to damp undesired vibrations resulting from [21] Zhao M, Pan JS, Chen ST. Optimal SNR of audio watermarking by wavelet and
seismic activities. According to the results, the proposed methodology is compact PSO methods. J Inform Hid Multimed Signal Process 2015;6(5):833–46.
[22] Chang FC, Huang HC. A refactoring method for cache-efficient swarm intelligence
suitable for optimally tuning TMDs. The results were compared with algorithms. Inf Sci 2012;192(1):39–49.
previously developed approaches employing harmony search, PSO and [23] Hadi MNS, Arfiadi Y. Optimum design of absorber for MDOF structures. J Struct
genetic algorithms. The proposed method is more effective than other Eng - ASCE 1998;124:1272–80.
[24] Singh MP, Singh S, Moreschi LM. Tuned mass dampers for response control of
approaches in finding better reductions for the maximum first story
torsional buildings. Earthq Eng Struct D 2002;31:749–69.
displacements. Additionally, the computational cost and reliability of [25] Desu NB, Deb SK, Dutta A. Coupled tuned mass dampers for control of coupled
the BA based method is better than HS based method. The optimum vibrations in asymmetric buildings. Struct Control Hlth 2006;13:897–916.
results are robust and not local optima. These conclusions were proved [26] Pourzeynali S, Lavasani HH, Modarayi AH. Active control of high rise building
structures using fuzzy logic and genetic algorithms. Eng Struct 2007;29:346–57.
with additional analyses after the optimization process by changing the [27] Marano GC, Greco R, Chiaia B. A comparison between different optimization cri-
stiffness of the structure and using extra benchmark earthquakes which teria for tuned mass dampers design. J Sound Vib 2010;329:4880–90.
are not considered in the optimization process. As a final conclusion, BA [28] Steinbuch R. Bionic optimisation of the earthquake resistance of high buildings by
tuned mass dampers. J Bionic Eng 2011;8:335–44.
employed TMD optimization methodology is the best existing ap- [29] Leung AYT, Zhang H. Particle swarm optimization of tuned mass dampers. Eng
proach. Struct 2009;31:715–28.
It can be expected that the proposed approach can be used for other [30] Leung AYT, Zhang H, Cheng CC, Lee YY. Particle swarm optimization of TMD by
non-stationary base excitation during earthquake. Earthq Eng Struct D
design optimization problems in other applications related to seismic 2008;37:1223–46.
engineering. Further research topics can extend the current method to [31] Bekdaş G, Nigdeli SM. Estimating optimum parameters of tuned mass dampers
deal with structures with uncertainty and noise, and to solve large-scale using harmony search. Eng Struct 2011;33:2716–23.
[32] Bekdaş G, Nigdeli SM. Optimization of tuned mass damper with harmony search. In:
optimization problems.
Gandomi Amir Hossein, Yang Xin-She, Alavi Amir Hossein, Talatahari Siamak,
editors. Metaheuristic applications in structures and infrastructures, Elsevier,
References February 2013 [chapter 14].
[33] Bekdaş G, Nigdeli SM. Mass ratio factor for optimum tuned mass damper strategies.
Int J Mech Sci 2013;71:68–84.
[1] Frahm H. Device for damping of bodies. U.S. Patent No: 989,958; 1911. [34] Zhang HY, Zhang LJ. Tuned mass damper system of high-rise intake towers opti-
[2] Ormondroyd J, Den Hartog JP. The theory of dynamic vibration absorber. T. ASME mized by improved harmony search algorithm. Eng Struct 2017;138:270–82.
1928;50:9–22. [35] Yang X-S. A new metaheuristic bat-inspired algorithm. In: Gonzalez JR et al., edi-
[3] Den Hartog JP. Mechanical vibrations. 3rd ed. New York: Mc Graw-Hill; 1947. tors. Nature Inspired Cooperative Strategies for Optimization (NISCO 2010), Studies
[4] Warburton GB, Ayorinde AE. Optimum absorber parameters for simple systems. in Computational Intelligence, Springer, Berlin, 284, Springer, 65–74; 2010.
Earthq Eng Struct D 1980;8:197–217. [36] Yang XS, He X. Bat algorithm: literature review and applications. Int J Bio-Inspir
[5] Warburton GB. Optimum absorber parameters for various combinations of response Comput 2013;5(3):141–9.
and excitation parameters. Earthq Eng Struct D 1982;10:381–401. [37] The MathWorks Inc. MATLAB R2010a. Natick, MA,USA; 2010.
[6] Rana R. A parametric study of tuned mass dampers and their generalizations, M.S. [38] Geem ZW, Kim JH, Loganathan GV. A new heuristic optimization algorithm: har-
thesis, State University of New York, Bufallo; 1995. mony search. Simulation 2001;76:60–8.
[7] Rana R, Soong TT. Parametric study and simplified design of tuned mass dampers. [39] Pacific Earthquake Engineering Research Center (PEER NGA DATABASE) < http://
Eng Struct 1998;20:193–204. peer.berkeley.edu/nga > .
[8] Sadek F, Mohraz B, Taylor AW, Chung RM. A method of estimating the parameters [40] Liu MY, Chiang WL, Hwang JH, Chu CR. Wind-induced vibration of high-rise
of tuned mass dampers for seismic applications. Earthq Eng Struct D building with tuned mass damper including soil–structure interaction. J Wind Eng
1997;26:617–35. Ind Aerodyn 2008;96:1092–102.
[9] Chang CC. Mass dampers and their optimal designs for building vibration control.
Eng Struct 1999;21:454–63.

98