Sei sulla pagina 1di 11

Journal of Sound and Vibration 335 (2015) 55–65

Contents lists available at ScienceDirect

Journal of Sound and Vibration


journal homepage: www.elsevier.com/locate/jsvi

H1 optimization of dynamic vibration absorber variant


for vibration control of damped linear systems
Semin Chun, Youngil Lee, Tae-Hyoung Kim n
School of Mechanical Engineering, College of Engineering, Chung-Ang University, 221 Heukseok-dong, Dongjak-gu, Seoul 156-756,
Republic of Korea

a r t i c l e i n f o abstract

Article history: This study focuses on the H 1 optimal design of a dynamic vibration absorber (DVA)
Received 23 December 2013 variant for suppressing high-amplitude vibrations of damped primary systems. Unlike
Received in revised form traditional DVA configurations, the damping element in this type of DVA is connected
5 August 2014
directly to the ground instead of the primary mass. First, a thorough graphical analysis of
Accepted 17 September 2014
the variations in the maximum amplitude magnification factor depending on two design
Handling Editor: D.J. Wagg
Available online 18 October 2014 parameters, natural frequency and absorber damping ratios, is performed. The results of
this analysis clearly show that any fixed-points-theory-based conventional method could
provide, at best, only locally but not globally optimal parameters. Second, for directly
handling the H 1 optimization for its optimal design, a novel meta-heuristic search engine,
called the diversity-guided cyclic-network-topology-based constrained particle swarm
optimization (Div-CNT-CPSO), is developed. The variant DVA system developed using the
proposed Div-CNT-CPSO scheme is compared with those reported in the literature. The
results of this comparison verified that the proposed system is better than the existing
methods for suppressing the steady-state vibration amplitude of a controlled primary
system.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Dynamic vibration absorbers (DVAs) are widely used as passive vibration control devices by attaching them to a vibrating
system, called the primary system. The first DVA invented [1] was a mass-spring system without damping element. When
correctly tuned and attached to a primary system subject to a harmonic exciting force, this device would stop steady-state
motion at the point to which it was attached [2]. It was a simple vibration suppression solution, but useful only in a narrow
range of frequencies close to the DVA's natural frequency [3]. Therefore, Ormondroyd and Den Hartog [4] proposed the
so-called Voigt-type DVA in which both a spring element and a damping element are arranged in parallel between the
absorber mass and the primary mass. This mechanism is effective over an extended range of frequencies, but cannot
completely eliminate steady-state vibrations of the primary system. From this viewpoint, design optimization of the Voigt-
type DVA is considered important in the field of vibrations.
Four model configurations of systems with a damped DVA are shown in Fig. 1: (Case 1) T-DVA-UPS: traditional DVA
attached to an undamped primary system, (Case 2) T-DVA-DPS: traditional DVA attached to a damped primary system, (Case 3)
V-DVA-UPS: variant DVA attached to an undamped primary system, and (Case 4) V-DVA-DPS: variant DVA attached to

n
Corresponding author.
E-mail addresses: chunsm@cau.ac.kr (S. Chun), yilee@cau.ac.kr (Y. Lee), kimth@cau.ac.kr (T.-H. Kim).

http://dx.doi.org/10.1016/j.jsv.2014.09.020
0022-460X/& 2014 Elsevier Ltd. All rights reserved.
56 S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65

Fig. 1. Four types of damped DVAs for damped/undamped primary systems and a few related researches.

a damped primary system. In Cases 1 (T-DVA-UPS) and 2 (T-DVA-DPS), Voigt-type DVAs are directly attached to undamped and
damped primary systems, respectively. In contrast, in Cases 3 (V-DVA-UPS) and 4 (V-DVA-DPS), the damping element connects
the absorber mass directly to the ground, unlike the traditional damped DVA configurations of Cases 1 and 2. In the following
text, existing literatures are briefly reviewed.
(A) Case 1 (T-DVA-UPS): In 1932, Hahnkamm [5] first deduced that all frequency response curves of the undamped
primary system pass through two invariant points independent of absorber damping when a harmonic force is applied to
the primary mass. Following this characteristic, the desired optimum value of the absorber's resonance frequency can be
found when the heights of two points are equal. In 1946, Brock [6] developed the mathematical method known as fixed-
points theory and showed that the optimum absorber damping can be detected by making the height of the fixed points the
maximum. However, the above-mentioned examples are not exact but approximate solutions for the H 1 optimization of
DVAs because some approximations were made in deriving these solutions [2,3]. In contrast, Nishihara and Matsuhisa [7]
presented an exact solution for the H 1 optimal design of T-DVA-UPS.
(B) Case 2 (T-DVA-DPS): The fixed-points theory has been adopted as the fundamental tool for analytically determining
the optimum parameters of a traditional DVA attached to a primary system without damping. However, this theory does not
hold in the case that a damping element is introduced in the primary system because no fixed points exist in the frequency
response curves [3]. Given this background, most researchers working on the optimal design of this case have focused on
developing efficient numerical approaches [8–14]. Unlike the above approaches, Asami et al. [3] used a perturbation method
and then derived a series solution in the form of a power series in powers of the primary damping ratio. However, their
analytical expressions for the optimal parameters are lengthy and complex, and may not be applied easily in practice [2].
(C) Case 3 (V-DVA-UPS): This variant of the DVA was proposed by Ren [15]. Its optimal design was then studied by Liu and
Liu [16], but an erroneous DVA damping ratio expression was presented [17]. Notably, the conventional studies of Ren [15],
Liu and Liu [16], Wong and Cheung [18] introduced the fixed-points theory for determining the optimum tuning parameters
for minimizing the resonant vibration of this type of system. A recent study by Cheung and Wong [2], however, proved that
such optimum tuning parameters do not lead to the minimum resonant amplitude of a single degree-of-freedom system
subject to harmonic force excitation. Then, they proposed a new procedure for the H 1 optimization of such a DVA, which
yields lower maximum amplitude responses than those reported in the literatures.
(D) Case 4 (V-DVA-DPS): Similar to Case 2 (T-DVA-DPS), it is considerably difficult to find the exact analytical solutions for
the optimal parameters of a variant DVA combined with a damped primary system because the fixed-points theory does not
hold. Therefore, few studies have been reported on this optimal design problem. Liu and Coppola [19] derived approximate
optimum tuning parameters analytically, and then employed two numerical methods to find those parameters. The first
numerical approach involved using the Chebyshev's equioscillation theorem, and the second theory involved using the
Nelder–Mead or the sequential simplex method. However, their approaches were mainly based on the fixed-points theory
with a somewhat vague assumption that a damped DVA with a small mass ratio is attached to a lightly or moderately
damped primary system. Anh and Nguyen [17] very recently proposed an approach for determining approximate analytical
solutions for H 1 optimization. They first introduced the dual equivalent linearization technique for approximately replacing
S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65 57

the original damped primary system with an equivalent undamped system, and then derived the optimum natural
frequency ratio. However, another design variable, absorber damping ratio, was set equal to that of the derived undamped
primary system by using the fixed-points theory, which was justified based only on the empirical observation that variation
in the maximum amplitude magnification factor may be less sensitive to changes in the damping ratio. However, the above
studies provide no thorough analysis of the system's vibration suppression characteristics with respect to variations in two
design variables—natural frequency and absorber damping ratio. Furthermore, the design variables derived based on the
methods of Liu and Coppola [19] and Anh and Nguyen [17] may not result in the minimum resonant vibration amplitude of
the primary mass.
This study analyzes the system of Case 4 (V-DVA-DPS) with a focus on (i) analyzing the distinctive characteristics of the
maximum of amplitude magnification factor of a primary system, and (ii) developing direct method for attacking its H 1
optimal design problem. To this end, we first briefly revisit the conventional studies of Liu and Coppola [19] and Anh and
Nguyen [17] and then show that their schemes providing two design variables, natural frequency and absorber damping
ratio, may not effectively minimize the maximum vibration amplitude of a damped primary mass. In fact, their approximate/
numerical solution methods were developed by applying the fixed-points theory both directly and indirectly to a system
(V-DVA-DPS) for which the theory simply does not hold any longer. Consequently, the designed variables could not be the
optima in the strictest sense. For verifying the above fact, variations in the maximum amplitude magnification factor were
thoroughly analyzed as functions of the natural frequency and absorber damping ratio. This graphical analysis clearly shows
that any conventional method could provide, at best, only locally but not globally optimal solutions. In contrast, the fact that
the variation trend of the maximum amplitude magnification factor depends on not only two design variables but also the
mass ratio complicates systematic determination of the globally optimum set of natural frequencies and absorber damping
ratios. To overcome this difficulty readily, we propose diversity-guided cyclic-network-topology-based constrained particle
swarm optimization (Div-CNT-CPSO), which directly solves the H 1 optimization for a variant form of a DVA attached to
a damped primary system. This alternative method directly handles the optimal design of a variant DVA model in Case 4
(V-DVA-DPS) without introducing considering the fixed-points theory. Therefore, most practical engineers can use it
without any difficulty. Furthermore, because the proposed method considers DVA structure in a straightforward manner in
the overall system design, the methods can be applied to the design of any type of DVA, as shown in Fig. 1 so long as a
suitable formulation for the amplitude magnification factor is derived. The developed optimal variant DVA system is
compared with those of Liu and Coppola [19] and Anh and Nguyen [17], and it is verified that the proposed system achieves
better suppression of the steady-state vibration amplitude of a controlled primary system.

2. Variant of DVA attached to damped primary system

2.1. System description and problem formulation

The variant DVA model considered in this paper is Case 4 (V-DVA-DPS) of Fig. 1. The dynamics of the primary mass (M)
subject to a sinusoidal excitation and the absorber mass (m) are given as follows:
M x€ p þC x_ p þKxp þ kðxp xa Þ ¼ F sin ðωtÞ; (1)

mx€ a þcx_ a þkðxa xp Þ ¼ 0; (2)


where M, C, and K are the mass, damping, and stiffness value, respectively, of the primary system; m, c, and k are the mass,
damping, and stiffness value of the absorber system; xp and xa are the displacements of the primary mass and the absorber
mass; and F and ω are the amplitude and frequency of the exciting force. Then, the amplitude magnification factor of the
primary mass can be written as follows:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
  u
   xp  u ðν2  λ Þ2 þ ð2ξa νλÞ2
2
G ν; ξa ≔ ¼t ; (3)

½λ  ð1 þ ν þ μν þ4ξ ξ νÞλ þ ν 2 þ 4½ðξ ν þ ξ Þλ  ðξ þ μξ ν2 þ ξ νÞνλ2
F=K 4 2 2 2 2 3
a p a p a a p

where ν is the natural frequency ratio, λ is the forced frequency ratio, μ is the mass ratio, ξa is the absorber damping ratio,
and ξp is the primary damping ratio. These are defined as follows:
pffiffiffiffiffiffiffiffiffi
k=m ω m c C
ν ¼ pffiffiffiffiffiffiffiffiffiffiffi; λ ¼ pffiffiffiffiffiffiffiffiffiffiffi; μ ¼ ; ξa ¼ pffiffiffiffiffiffiffi; ξp ¼ pffiffiffiffiffiffiffiffi: (4)
K=M K=M M 2 mk 2 MK
For a given μ and ξp, the objective of H 1 optimization is to find two design variables, ν and ξa (i.e., k and c), that minimize
the H 1 norm (i.e., the maximum of Gðν; ξa Þ in (3)), which is formulated as follows:
 
min max Gðν; ξa Þ : (5)
ν;ξa

On the other hand, the fixed-points theory is not valid in this DVA system. For the purpose of illustration, frequency
response curves of the primary mass with μ ¼ 0.1, ξp ¼ 0:08, ν ¼ 1, and ξa ¼ 0:01, 0.1, 0.3, 0.5 are calculated using (3), and
the results are plotted in Fig. 2. These curves verify that the points through which all curves pass independent of the
absorber damping ratios do not exist. This fact implies that the conventional results of Liu and Coppola [19] and Anh and
58 S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65

6
ξ = 0.01
a
ξ = 0.1
a

Amplitude magnification factor (G ( ν,ξa ))


5
ξa= 0.3
ξa= 0.5
4

0
0 0.5 1 1.5 2
Forced frequency ratio (λ)

Fig. 2. Frequency response curves of primary mass with μ ¼ 0:1, ξp ¼ 0:08 and ν ¼ 1 at four different absorber damping ratios.

Nguyen [17], obtained using the fixed-points theory directly or indirectly, cannot generate the optimal values of ν and ξa.
A brief discussion on the results of Liu and Coppola [19] and Anh and Nguyen [17] is presented in the following section.

2.2. Brief discussion on conventional studies

For the variant DVA model in Case 4 (V-DVA-DPS) shown in Fig. 1, Liu and Coppola [19] first suggested a method for
analytically determining the optimum tuning parameters. They applied the fixed-points theory in doing so under a
somewhat vague assumption that this theory approximately holds for the case when the primary system is lightly or
moderately damped. Then, they showed that the minimum of the maximum of Gðν; ξa Þ in (3) becomes
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
minðmax GðνÞÞ ¼ ; (6)
4ξp λ þð1 þ ν2 μ  λ Þ2
2 2 2

where
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1  4ξp
2
1 þð1 þ μÞν2 7 1 þ2ðμ 1Þν2 þðμ2 þ 1Þν4
ν¼ ; λ¼ (7)
1μ 2

Note that (6) does not depend on ξa according to the fixed-points theory. However, the above solution is just an approximate
one because no fixed points exist, as shown in Fig. 2.
Liu and Coppola [19] also employed two numerical methods to this design problem. The first approach adopted the
Chebyshev's equioscillation theorem for finding the optimum values of ν and ξa such that the amplitude magnification
factor, Gðν; ξa Þ, has two equal peak values with a minimal distance from a certain straight line. Then, they showed that the
optimum design parameters can be obtained by solving the six derived nonlinear algebraic equations with seven unknowns;
this was achieved by prescribing a value for one of the seven unknowns. It implies that the introduced design procedure
may just provide local optima, not global optima. The second approach was based on the Nelder–Mead technique, which is a
commonly used heuristic nonlinear optimization scheme, for solving a weighted objective function with two criteria. The
first criterion is for minimizing the difference between two peak values, and the second criterion is for minimizing the
maximum of Gðν; ξa Þ. However, because the suggested approaches were derived mainly based on the inapplicable fixed-
points theory, we cannot assure that the best vibration suppression performance is achieved.
In contrast, Anh and Nguyen [17] recently adopted the dual equivalent linearization technique for handling the variant
DVA model of Case 4 (V-DVA-DPS), which transforms approximately the damped primary system to an equivalent
undamped system. Based on the above procedure, two design parameters were presented as follows:
1
ν¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !; (8)
pffiffiffiffiffiffiffiffiffiffiffi π2 π
1μ 1þ ξ2 þ ξp
ðπ 2 2Þ 2 p π2  2
sffiffiffiffiffiffiffiffiffiffiffi
1 3μ
ξa ¼ : (9)
2 2μ
S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65 59

2.5

Amplitude magnification factor (G( ν, ξ ))


a
2

1.5

0.5 ν = 1.8, ξ = 0.7001


a
Liu and Coppola [20]
Anh and Nguyen [18]
0
0 0.5 1 1.5 2
Forced frequency ratio ( λ)

Fig. 3. Comparison of amplitude magnification factors where μ ¼ 0:1 and ξp ¼ 0:08.

However, the absorber damping ratio in (9) is the same as that derived based on the fixed-points theory for a variant DVA
attached to an undamped primary mass (Case 3 (V-DVA-UPS), as shown in Fig. 1). Although they claimed its appropriacy on
the empirical basis that the variation of the maximum amplitude magnification factor may be less sensitive to changes in
the damping ratio, the presented design variables were not precisely the optimum variables.
The following example justifies why the design approaches of Liu and Coppola [19] and Anh and Nguyen [17] cannot
achieve the best performance for vibration suppression of the primary mass.

Example 1. This numerical example is taken from Anh and Nguyen [17], where the primary damping ratio (ξp) is 0.08 and
the mass ratio (μ) is 0.1. Two amplitude magnification factors corresponding to the tuning methods of Liu and Coppola [19]
and Anh and Nguyen [17] are drawn in Fig. 3. The dotted-dashed line is the result of Liu and Coppola [19] with ν ¼ 1.0405
and ξa ¼ 0:1987,1 and the dashed line is the result of Anh and Nguyen [17] with ν ¼ 1:021 and ξa ¼ 0:1987. The figure shows
that in terms of H 1 optimization, the design scheme of Anh and Nguyen [17] yields better vibration suppression result than
that of Liu and Coppola [19] in this case. However, the solid line corresponding to ν ¼ 1:8 and ξa ¼ 0:7001 surpasses the
vibration suppression performance of Anh and Nguyen [17]. This clearly verifies that the conventional tuning schemes
cannot achieve the optimum design of the variant DVA system in Case 4 (V-DVA-DPS), as shown in Fig. 1. Therefore, a
thorough analysis of the variant DVA's vibration suppression characteristics with respect to variations in the system
parameters in (4) is required, which is given in the following section.

2.3. Graphical analysis: changing pattern of maximum amplitude magnification factor

To analyze why the conventional tuning methods do not provide the optimum design of Example 1, we solve the H 1
optimization problem (5) for various prescribed natural frequency ratio values,2 i.e., find ξa minimizing max Gðν; ξa Þ for a
given μð ¼ 0:1Þ, ξp ð ¼ 0:08Þ, and ν. The upper panel of Fig. 4 shows the obtained optimal ξa corresponding to a prescribed ν
where the search range of λ was set as 0 r λ r 4. The lower panel shows the resulting max Gðν; ξa Þ. Note that, in this case, the
upper bound of ξa is set to 1ð Z ξa Z 0Þ.3 The designed values of ξa and ν by Liu and Coppola [19] and Anh and Nguyen [17] are
marked with a square and a circle, respectively, in Fig. 4 (upper), and the corresponding values of max Gðν; ξa Þ are marked in
Fig. 4 (lower). The point marked with a triangle in Fig. 4 (upper) denotes ðν; ξa Þ ¼ ð1:8; 0:7001Þ, which produces lower
max Gðν; ξa Þ than those of Liu and Coppola [19] and Anh and Nguyen [17], as can be verified from Figs. 3 and 4 (lower).
The analytical solution of Liu and Coppola [19] was developed based on the fixed-points theory, which produces a
frequency response curve of the primary mass with two peaks of identical values. Two other numerical methods also
adopted the same frequency response properties. Hence, their design parameters cannot provide the optimum design, as
shown in Fig. 4. Furthermore, the resulting amplitude magnification factor shown in Fig. 3 has two equal peaks. In addition,
the ν of Anh and Nguyen [17] was approximately derived and ξa was set equal to a previously obtained value without
considering the damping effect on the primary mass. Therefore, their method cannot be guaranteed to provide the best
vibration suppression performance, as verified in Fig. 4. Fig. 4 shows that there are alternatives of ðν; ξa Þ, which achieve more
desirable results, and ðν; ξa Þ ¼ ð1:8; 0:7001Þ is one such solution. Furthermore, the pairs of ðν; ξa Þ from Liu and Coppola [19]

1
These parameter values were presented in Anh and Nguyen [17].
2
This problem is solved by using the Div-CNT-CPSO algorithm presented in Section 3.2.
3
According to Cheung and Wong [2], this is a practical constraint that should be considered in the design formulation of a vibration system.
60 S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65

Optimal ξa
0.5

6
max G (ν, ξ )
a

4
ν = 1.8, ξ a= 0.7001
2 Liu and Coppola [20]
Anh and Nguyen [18]
0
0 0.5 1 1.5 2 2.5
Natural frequency ratio (ν)

Fig. 4. Graphical representation of optimal ξa corresponding to prescribed ν (upper figure), and resulting minimum value of max Gðν; ξa Þ (lower figure).

0.26
3.0
ν = 1.037, ξ a= 0.2315
0.24
Liu and Coppola [20]
Optimal ξa

0.22 Anh and Nguyen [18]


2.5
Amplitude magnification factor (G( ν, ξ ))
a
0.20
0.18 2.0

1.5
2.9
max G (ν, ξa )

1.0
2.8

2.7 0.5 ν = 1.037, ξa = 0.2315


Liu and Coppola [20]
Anh and Nguyen [18]
2.6 0
1 1.01 1.02 1.03 1.04 1.05 0 0.5 1 1.5 2

Natural frequency ratio (ν) Forced frequency ratio ( λ)

Fig. 5. Graphical analysis of optimal design with ðνopt ; ξa;opt Þ ¼ ð1:037; 0:2315Þ, and its comparison with two conventional methods. (a) Optimal ξa, and
corresponding max Gðν; ξa Þ. (b) Comparison of amplitude magnification factors.

1.0 1.0
Optimal ξa

Optimal ξa

0.5 0.5
Liu and Coppola [20]
Anh and Nguyen [18]
0 0

6.5
6 Liu and Coppola [20]
6.0
Anh and Nguyen [18]
max G (ν, ξ )

max G (ν, ξa )
a

5.5 4
5.0
2
4.5
4.0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2
Natural frequency ratio (ν) Natural frequency ratio (ν)

Fig. 6. Optimal ξa corresponding to prescribed ν (upper figure) and resulting minimum value of max Gðν; ξa Þ (lower figure). (a) Variant DVA model with
μ ¼ 0:01 and ξp ¼ 0:08. (b) Variant DVA model with μ ¼ 0:25 and ξp ¼ 0:08.

and Anh and Nguyen [17] were never the local optimum, which provides the minimum of max Gðν; ξa Þ even for
1:01 r ν r1:05. This is confirmed in Fig. 4 (lower). In fact, when the desired range of ν is locally limited to
1:0 r ν r 1:05, a quasi-optimal solution can be approximated as ðνn ; ξa Þ  ð1:037; 0:2315Þ, which is shown in Fig. 5.
n
S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65 61

On the other hand, if the mass ratio, μ, of the variant DVA model in Example 1 varies from 0.1 to 0.01 or 0.25, the optimal
ξa and the corresponding max Gðν; ξa Þ are obtained, as shown in Fig. 6(a) and (b), respectively. These observations show the
influence of mass ratio on the changing pattern of the maximum magnitude of Gðν; ξa Þ. Fig. 6(a), which corresponds to a
relatively small μ, shows that max Gðν; ξa Þ decreases more steeply near ν ¼ 1 than the case of Fig. 4, and Liu and Coppola [19]
seems to provide a solution close to the optimum in this vicinity. However, their DVA design is not optimum in over a
broader range of natural frequency ratios. Furthermore, if μ takes a relatively high value, max Gðν; ξa Þ continues to decrease
as ν increases, as shown in Fig. 6(b). Therefore, it is difficult to say in this case that the conventional methods of Liu and
Coppola [19] and Anh and Nguyen [17] provide even locally optimal solutions. In Figs. 4 and 6, note that the minimum of
max Gðν; ξa Þ can be achieved for a high value of ν irrespective of μ. However, such a natural frequency ratio ν may be too
high to be applied in practice, and there may be practical restrictions on realizing the corresponding absorber damping ratio
ξa [2].
The aforementioned analysis method could also be adopted for determining the optimal ν and ξa of a variant DVA
system. However, the procedure to find, one by one, the optimal ξa for a large number of prescribed values of ν may be quite
cumbersome and time consuming. Therefore, in practical vibration control application, it is crucial to find the optimal design
variables precisely in a fairly simple manner owing to limitations of available computer resource and the need for occasional
on-site parameter tuning. The following section focuses on a direct method for attacking such a problem.

3. H 1 optimization of variant DVA system: Div-CNT-CPSO-based approach

3.1. Formulation of optimal design problem

The design variables for Case 4 (V-DVA-DPS) of Fig. 1 are natural frequency ratio ν and absorber damping ratio ξa. The
min–max optimization considered in this paper determines the values of these design variables such that the maximum
objective function value is minimized over the domain of interest of the forced frequency ratio λ, where the objective
function is the amplitude magnification factor Gðν; ξa Þ in (3). The optimization problem considered in this paper then can be
formulated as follows:
 
min max Gðν; ξa Þ ; (10)
ν;ξa λL r λ r λH

where λL and λH are given and confine the search range of λ. On the other hand, in some applications the minimum and
maximum values of the design variables in the design process should be limited out of consideration for implementation
practicality. In such cases, the above min–max optimization problem includes constraint conditions on ν and ξa: νL r ν r νH
and ξa;L r ξa r ξa;H where νL , νH , ξa;L , and ξa;H are given. Then, it follows that
h1 ðνÞ≔νL  ν r 0; h2 ðνÞ≔ν  νH r 0; h3 ðξa Þ≔ξa;L  ξa r 0; h4 ðξa Þ≔ξa  ξa;H r 0: (11)

Note that, for a variant DVA attached to an undamped primary system (i.e., case 3 (V-DVA-UPS) in Fig. 1), Cheung and Wong [2]
introduced
ppractical constraints
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffi as ξa r ξa;H r1, 0 o μ r0:25, and k r K. If these constraints hold for Case 4 (V-DVA-DPS),
psuch
then ν ¼ ðk=KÞð1=μÞ r νH r 1=μ.

3.2. Div-CNT-CPSO-based design

The PSO algorithm is capable of finding the global optima in a variety of unconstrained engineering optimization
problems. However, an important factor facilitating feasible global search in this optimal DVA design is consideration of the
constraint conditions in (11) for calculating the fitness value of each particle and determining the optimal design-variable
vector xn ≔fνn ; ξa g. Our constraint-handling strategy combined with the Div-CNT-CPSO algorithm relies on a simple
n

transformation of (10) subject to (11) into an unconstrained optimization problem using a pseudo-function such as
Gv ðxÞ≔arctanfGðxÞg  π =2 [20]. Then, the modified unconstrained problem is formulated as follows:
(
hmax ðxÞ if hmax ðxÞ≔max½h1 ðxÞ; h2 ðxÞ; h3 ðxÞ; h4 ðxÞ Z0;
min GðxÞ≔ (12)
x A R2 Gv ðxÞ otherwise:

Note that because Gv ðxÞ o 0 for any x, the condition, Gv ðxÞ ohmax ðxÞ, holds for an infeasible solution x satisfying hmax ðxÞ Z0.
Therefore, the modified problem ensures that any particle randomly positioned in an infeasible region of the search space
tries to move to a feasible region with a lower cost function value from the beginning of the optimization process. Then, the
Div-CNT-CPSO algorithm based on the modified unconstrained optimization of (12) proceeds as follows.
Step 1. Initialization of Div-CNT-CPSO parameters and commencement of optimization:

Let D denote the hyperdimensional search space of the design vector xℓi ≔fνℓ ; ξa g A R2 , where i ¼ 1; 2; …; np is the particle
index, ℓ ( ¼ 0; 1; …; ℓmax ) is the iteration number, and ℓmax is the maximum number of searches. Initialize np particles with
randomly selected positions x0i ≔fν0 ; ξa g A D and velocities v0i ¼ 0 A R2 . Let xℓpbest;i denote each particle's best previous
0

position, i.e., the position that yields the minimum fitness value of the modified cost function GðÞ in (12). Its initial value is
x0pbest;i ¼ x0i . Next, we introduce xℓsbest;i , which denotes the best position in the social neighborhood of the ith particle in the
62 S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65

current iteration ℓ. Mathematically, this can be written as follows:


xℓsbest;i ≔arg min GðxÞ; (13)
x A fxℓj jj ¼ i  ðns =2Þ;…;i þ ðns =2Þg

where ‘arg min’ denotes the set of points at which the cost function GðÞ attains its minimum value, (even) ns ð r np Þ denotes
the number of neighbors of the ith particle, and xℓj ≔xℓðj  1mod np Þ þ 1 for j o1 or np þ1 rj. Therefore, the initial x0sbest;i is set
using a previously determined ns as x0sbest;i ≔arg minx A fx0 jj ¼ i  ðns =2Þ;…;i þ ðns =2Þg GðxÞ.
j
Step 2. Apply the Div-CNT-CPSO algorithm for updating the positions xℓi and velocities vℓi of all particles:
0 0 0 0
After initializing xi , vi , xpbest;i , and xsbest;i in Step 1, the following velocity update law is applied to all particles, for
i ¼ 1; 2; …; np :
8
>
>
χ ½vℓi þ c1 rℓ1;i ðxℓpbest;i  xℓi Þ þ c2 rℓ2;i ðxℓsbest;i  xℓi Þ if DIVi ðℓÞ 4 Dhigh ;
<
vℓi þ 1 ’ χ ½vi  c1 r 1;i ðxpbest;i  xi Þ  c2 r 2;i ðxsbest;i  xi Þ if DIVi ðℓÞ o Dlow ;
ℓ ℓ ℓ ℓ ℓ ℓ ℓ
(14)
>
>
: χ ½vℓ þ c1 r ℓ ðxℓ  xℓ Þ  c2 r ℓ ðxℓ  xℓ Þ otherwise;
i 1;i pbest;i i 2;i sbest;i i

where the diversity measure of the ith particle, DIVi ðℓÞ, is defined as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !
1 i þ ðns =2Þ 2  1 i þ ðns =2Þ
ℓ 2 ℓ
DIVi ðℓÞ ¼ ∑ ∑ xℓp;q  x^ q ; x^ q ≔ ∑ xℓr;q : (15)
ns þ 1 p ¼ i  ðns =2Þ q ¼ 1 ns þ 1 r ¼ i  ðns =2Þ

Here, Dhigh and Dlow are selected by the designer, r ℓ1;i and r ℓ2;i denote two random numbers generated uniformly within ½0; 1,
c1 is the cognitive scaling factor, c2 is the social scaling factor, χ is the constriction factor defined as
2
χ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (16)
j2  φ  φ2  4φj
for φ≔c1 þ c2 ð 44Þ, and xp;q denotes the qth entry of the pth particle's position vector. The above law states that the velocity
behavior of each particle is switched alternately to one of the phases of attraction, repulsion, and intermediate, based on an
increase or decrease, respectively, in diversity among neighboring particles during each iteration.4 The position of the ith
particle is then updated as follows:

xℓi þ 1 ’xℓi þ vℓi þ 1 : (17)

Next, we set ℓ’ℓ þ1 and update xℓpbest;i and xℓsbest;i as follows:


xℓpbest;i ’arg min GðxÞ; xℓsbest;i ’arg min GðxÞ: (18)
x A fxji jj ¼ 1;2;…;ℓg x A fxℓj jj ¼ i  ðns =2Þ;…;i þ ðns =2Þg

Note that the velocity update law in (14) is a variant of the form used in Maruta et al. [20]. Their research combined xℓsbest;i in
(13) with the conventional PSO method, but did not introduce the notion of diversity measure, i.e., their velocity update law
was as follows:

vℓi þ 1 ’χ ½vℓi þ c1 r ℓ1;i ðxℓpbest;i  xℓi Þ þ c2 r ℓ2;i ðxℓsbest;i xℓi Þ: (19)

Compared with Maruta et al. [20], our PSO scheme includes two mechanisms—a diversity-guided velocity update law in (14)
and restricted social best searching in (13) based on the cyclic network topology—for improving particle diversification,
which allows different regions of the design space to be explored efficiently and, thus, increases the anti-premature ability.
However, because the update of vℓi is governed by the switching rule, the intensification ability, which allows a fine-search
around the best region to refine and select the best candidate solution, may be weaker than (19). Therefore, for maintaining
an adequate balance between diversification and intensification, xℓi is updated as follows:

(i) If ℓ r ℓDIV , where ℓDIV is given by the designer, xℓi is updated by (17) with (14).
(ii) Otherwise, xℓi is updated by (17) with (19).

Step 3. Check the optimization process stop condition:


If the user-defined termination criterion (e.g., ℓ 4 ℓmax ) of the optimization process is satisfied, the iteration of swarm
movements following the Div-CNT-CPSO algorithm stops, and the final optimal solution is determined as follows:

xn ð ¼ fνn ; ξa gÞ≔arg
n
min GðxÞ: (20)
x A fxji ji ¼ 1;2;…;np ;j ¼ 1;2;…;ℓg

If the termination criterion is not satisfied, return to Step 2.

4
A similar notion of the diversity measure was proposed by Pant et al. [21]. However, their method used position data from the entire swarm, and the
behavior of all particles was governed by an identical diversity measure as a result.
S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65 63

The overall performance of our Div-CNT-CPSO method in the design of the variant DVA system of Case 4 (V-DVA-DPS),
shown in Fig. 1, is evaluated in the next section.

3.3. Numerical simulation and comparisons

The first numerical design for finding the optimal design variables, νn and ξa , was developed under the condition that
n

μ ¼ 0:1 and ξp ¼ 0:08. It was assumed that the upper and lower bounds of the design variables are given as 0 o ν r1:5 and
0 o ξa r 0:8, respectively. The user parameters of PSO were set as np ¼ 20, ns ¼ 4, c1 ¼ c2 ¼ 2:05, and ℓmax ¼ 300.
A comparison of the best solution ðνn ; ξa Þ and the corresponding maxðGðνn ; ξa ÞÞ with those obtained using other
n n

conventional methods [17,19] is summarized in Table 1. We can see that the proposed approach produces a DVA design
identical to that obtained following the numerical method of Liu and Coppola [19]. The amplitude magnification factors
determined using the design variable values listed in Table 1 are calculated and plotted in Fig. 7(a). These imply that our
optimal design is achieved naturally when the frequency response curve of the primary mass has two equal maximum
amplitude, and, hence, becomes identical to Liu and Coppola's [19] solution, who artificially introduced the fixed-points
theory in their design process. In fact, the ν  ξa  maxðGðν; ξa ÞÞ relationship exhibits a trend similar to that shown in Fig. 4.
This verifies that the applied constraint condition on ν, 0 o ν r 1:5, ensures that the minimum of maxðGðν; ξa ÞÞ is achieved
close to the first lower peak in Fig. 4 (lower). Therefore, other conventional approaches could also achieve
pffiffiffiffiffiffiffiffi
ffi similar vibration
suppression characteristics in this case. However, if the upper bound of ν increases to 3:1623ð ¼ 1=μÞ (i.e., 0 o ν r 3:1623),
the resulting low peak is no longer the minimum of maxðGðν; ξa ÞÞ, as shown in Fig. 7(b) and summarized in Table 2. Note

Table 1
Comparison of optimization results with the literature values where μ ¼ 0:1 and ξp ¼ 0:08: case of 0o ν r 1:5.

Variables/ Liu and Coppola [19] Liu and Coppola [19] Anh and Nguyen [17] This study (Div-CNT-CPSO-
function (approximate solution) (numerical solution) (Approximate solution) based approach)

νn 1.0405 1.0364 1.0210 1.0364


ξna 0.1987 0.2302 0.1987 0.2302

maxðGðνn ; ξna ÞÞ 2.7051 2.6224 2.6775 2.6224

3.0 3.0

2.5 2.5
Amplitude magnification factor (G( ν, ξ ))
Amplitude magnification factor (G( ν, ξ ))

a
a

2.0 2.0

1.5 1.5

1.0 1.0

This study This study


0.5 Liu and Coppola [20] (Numerical solution) 0.5 Liu and Coppola [20] (Numerical solution)
Liu and Coppola [20] (Approximate solution) Liu and Coppola [20] (Approximate solution)
Anh and Nguyen [18] Anh and Nguyen [18]
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Forced frequency ratio ( λ) Forced frequency ratio ( λ)

Fig. 7. Comparisons of amplitude magnification factors where μ ¼ 0:1 and ξp ¼ 0:08. (a) Case of 0 o ν r 1:5. (b) Case of 0o ν r 3:1623.

Table 2
Comparison of optimization results with the literature where μ ¼ 0:1 and ξp ¼ 0:08: Case of 0o ν r 3:1623.

Variables/ Liu and Coppola [19] Liu and Coppola [19] Anh and Nguyen [17] This study (Div-CNT-CPSO-
function (approximate solution) (numerical solution) (approximate solution) based approach)

νn 1.0405 1.0364 1.0210 3.1623


ξna 0.1987 0.2302 0.1987 0.8

maxðGðνn ; ξna ÞÞ 2.7051 2.6224 2.6775 1.5365


64 S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65

Table 3
Comparison of optimization results with the literature values where μ ¼ 0:25 and ξp ¼ 0:3.

Variables/ Liu and Coppola [19] Liu and Coppola [19] Anh and Nguyen [17] This study (Div-CNT-CPSO-
function (approximate solution) (numerical solution) (approximate solution) based approach)

νn 0.9238 1.1467 1.0247 1.3


ξna 0.3273 0.6295 0.3273 0.7298

maxðGðνn ; ξna ÞÞ 1.2270 1.0831 1.2421 1.0244

1.4

1.2
Amplitude magnification factor (G( ν, ξ ))
a

1.0

0.8

0.6

0.4
This study
Liu and Coppola [20] (Numerical solution)
0.2 Liu and Coppola [20] (Approximate solution)
Anh and Nguyen [18]

0
0 0.5 1 1.5 2
Forced frequency ratio ( λ)

Fig. 8. Comparison of amplitude magnification factors where μ ¼ 0:25 and ξp ¼ 0:3.

that the approximate solution methods of Anh and Nguyen [17] and Liu and Coppola [19] do not explicitly incorporate the
bound conditions for ν and ξa in the design process, and may thus not lead to the optimal design unlike the developed
method. Further, the numerical method of Liu and Coppola [19] requires that the amplitude magnification factors Gðν; ξa Þ
have two peaks for application of the equioscillation theorem. This may degrade the vibration suppression performance, as
verified in this simulation. Contrary to the above conventional methods, our Div-CNT-CPSO-based approach provides a
unified design tool that achieves the H 1 optimal design in any given situation.
The second simulation example targets the optimal design of a variant DVA system where μ ¼ 0:25, ξp ¼ 0:3, 0 o ν r 1:3
and 0 o ξa r0:8. This is considered mainly to show that the fixed-points theory–based numerical approach of Liu and
Coppola [19], developed under the assumption that a damped DVA with a small mass ratio is attached to a lightly
or moderately damped primary system, cannot produce a reliable optimum design. Note that the trend of the
ν–ξa–maxðGðν; ξa ÞÞ relationship is similar to that shown in Fig. 6(b). A comparison of the design results obtained following
Liu and Coppola [19], Anh and Nguyen [17], and our approach is presented in Table 3 and shown in Fig. 8. It can be easily
verified that the variables designed with the approximate solution methods of Liu and Coppola [19] and Anh and Nguyen [17]
are far from the optimal ones. Furthermore, the optimal solution obtained using the numerical method of Liu and Coppola [19]
flattens the amplitude magnification factor curve owing to the increased μ and ξp, as mentioned in their paper, but it also cannot
achieve better vibration suppression characteristics than our optimum design result.
The two above simulation examples verify that the proposed design approach provides a promising alternative method
for directly arriving at the optimal design of the variant DVA model in Case 4 (V-DVA-DPS), shown in Fig. 1, without having
to consider the fixed-points theory. Therefore, most practicing engineers can use the method without any difficulty for
finding optimal DVA parameters that satisfy the various design constraints introduced. Furthermore, because our method
considers DVA structure in a straightforward manner in the overall system design, it can be applied to design problems
pertaining to the DVA types shown in Fig. 1 so long as a suitable formulation of the amplitude magnification factor is
derived.

4. Conclusion

In this article, the H 1 optimal design of a DVA variant was studied from the viewpoint of suppressing high-amplitude
vibrations of damped primary systems. The first step toward achieving this aim was based on a thorough graphical analysis
of variations in the maximum amplitude magnification factor depending on natural frequency and absorber damping ratios.
S. Chun et al. / Journal of Sound and Vibration 335 (2015) 55–65 65

This analysis clearly verified that any fixed-points-theory-based conventional method could provide, at best, only locally but
not globally optimal parameters. The second step was the development of a diversity-guided cyclic-network-topology-based
constrained particle swarm optimization (Div-CNT-CPSO) that directly performs the H 1 optimization of such a DVA variant
attached to a damped primary system. This scheme does not consider the fixed-points theory and allows for precisely
determining the optimal design variables within the implementable practical limits of their minimum and maximum values
in a fairly simple manner. Furthermore, since the Div-CNT-CPSO-based design method is straightforward in considering the
structure of a DVA in the overall system design, it can be applied to the design of any type of DVA. The variant DVA system
developed based on the proposed Div-CNT-CPSO scheme was compared with those reported in the literature. It was verified
that the proposed system is the most effective for suppressing the vibration of a controlled primary mass.

Acknowledgments

This research was supported by Basic Science Research Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Education, Science and Technology (No. 2012-012295), and the Chung-Ang University
excellent student scholarship in 2014.

References

[1] H. Frahm, Device for Damping Vibrations of Bodies, U.S. Patent, No. 989, 958 (1911) 3576–3580.
[2] Y.L. Cheung, W.O. Wong, H-infinity optimization of a variant design of the dynamic vibration absorber – revisited and new results, Journal of Sound and
Vibration 330 (16) (2011) 3901–3912.
[3] T. Asami, O. Nishihara, A.M. Baz, Analytical solution to H1 and H2 optimization of dynamic vibration absorbers attached to damped linear systems,
Journal of Vibration and Acoustics 124 (2) (2002) 284–295.
[4] J. Ormondroyd, J.P. Den Hartog, Thee theory of the dynamic vibration absorber, Journal of Applied Mechanics 50 (7) (1928) 9–22.
[5] E. Hahnkamm, Die Dämpfung von Fundamentschwingungen bei veränderlicher Erregergrequenz, Ingenieur Archive 4 (1932) 192–201. (in German).
[6] J.E. Brock, A note on the damped vibration absorber, ASME Journal of Applied Mechanics 13 (4) (1946)., p. A-284.
[7] O. Nishihara, H. Matsuhisa, Design and tuning of vibration control devices via stability criterion, The Japan Society of Mechanical Engineers 97 (10–1)
(1997) 165–168.
[8] K. Ikeda, T. Ioi, On the dynamic vibration damped absorber of the vibration system, Bulletin JSME 21 (151) (1978) 64–71.
[9] S.E. Randall, D.M. Halsted, D.L. Taylor, Optimum vibration absorbers for linear damped systems, ASME Journal of Mechanical Design 103 (4) (1981)
908–913.
[10] A.G. Thompson, Optimum tuning and damping of a dynamic vibration absorber applied to a forced excited and damped primary system, Journal of
Sound and Vibration 77 (3) (1981) 403–415.
[11] A. Soom, M. Lee, Optimal design of linear and nonlinear vibration absorbers for damped systems, ASME Journal of Vibration Acoustics 105 (1) (1983)
112–119.
[12] H. Sekiguchi, T. Asami, Theory of vibration isolation of a system with two degree of freedom, Bulletin JSME 27 (234) (1984) 2839–2846.
[13] E. Pennestrì, An application of Chebyshev's min–max criterion to the optimal design of a damped dynamic vibration absorber, Journal of Sound and
Vibration 217 (4) (1998) 757–765.
[14] O. Nishihara, T. Asami, Closed-form solutions to the exact optimizations of dynamic vibration absorber (minimizations of the maximum amplitude
magnification factors), Journal of Vibration and Acoustics 124 (2002) 576–582.
[15] M.Z. Ren, A variant design of the dynamic vibration absorber, Journal of Sound and Vibration 245 (4) (2001) 762–770.
[16] K. Liu, J. Liu, The damped dynamic vibration absorber: revisited and new result, Journal of Sound and Vibration 284 (2005) 1181–1189.
[17] N.D. Anh, N.X. Nguyen, Design of non-traditional dynamic vibration absorber for damped linear structures, Proceedings of the Institution of Mechanical
Engineers, Part C: Journal of Mechanical Engineering Science 228 (1) (2014) 45–55.
[18] W.O. Wong, Y.L. Cheung, Optimal design of a damped dynamic vibration absorber for vibration control of structure excited by ground motion,
Engineering Structures 30 (2008) 282–286.
[19] K. Liu, G. Coppola, Optimal design of damped dynamic vibration absorber for damped primary systems, Transactions of the Canadian Society for
Mechanical Engineering 34 (1) (2010) 119–135.
[20] I. Maruta, T.-H. Kim, D. Song, T. Sugie, Synthesis of fixed-structure robust controllers using a constrained particle swarm optimizer with cyclic
neighborhood topology, Expert systems with applications 40 (9) (2013) 3595–3605.
[21] M. Pant, T. Radha, V.P. Singh, A simple diversity guided particle swarm optimization, IEEE Congress on Evolutionary Computation, 2007, pp. 3294–3299.

Potrebbero piacerti anche