Engineering Structures 28 (2006) 1774–1779

www.elsevier.com/locate/engstruct

Tuned mass damper with fractional derivative damping

∗
F. Rudinger¨
Department of Mechanical Engineering, Technical University of Denmark, DK-2800 Kgs. Lyngby, Denmark

Received 29 March 2005; received in revised form 12 January 2006; accepted 16 January 2006
Available online 4 August 2006

Abstract

A tuned mass damper with a viscoelastic damping element applied to a single-degree-of-freedom system excited by white noise is considered. The
viscoelastic damping is modelled as a force proportional to the fractional derivative of the relative displacement between the structure and the secondary
mass. Optimal parameters for the tuned mass damper are obtained numerically by optimizing the effective damping ratio of the system. It is shown that
the structural damping has very little influence on the optimal parameters. Furthermore, it is demonstrated that the effect of the damper is the same for
different values of the fraction in the fractional derivative. This implies that this tuned mass damper with a fractional derivative damping element
introduces the same reduction in the structural vibration as a conventional tuned mass damper if properly tuned. Simple approximate analytical
expressions for optimal parameters are obtained by a frequency domain approach, in which the force acting between the structure and the secondary
mass is assumed to be equal to the force of a conventional tuned mass damper at resonance.
c 2006 Published by Elsevier Ltd

Keywords: Tuned mass damper; Fractional derivative damping; Optimal control; Structural dynamics; Passive damping; Random vibration

1. Introduction commercially available dampers or damping materials show a

behaviour, which deviates substantially from that of a linear
Tuned mass dampers offer a relatively simple and effective viscous damper. Different models for viscoelastic behaviour
way of reducing excessive vibrations of high rise buildings, applying the concept of fractional derivatives are discussed by
towers and chimneys. By attaching a secondary mass to the e.g. Bagley and Torvik [3,4].
structure with approximately the same natural frequency, large In the following analysis it is assumed that the damping
relative displacements between the structure and the force is proportional to a fractional derivative of the relative
secondary mass will occur at resonance, and the mechanical displacement between the structure and the secondary mass.
energy of the system can then be dissipated by placing a The structure is assumed to be excited by white noise. The
properly tuned damper between the two. If the fixed points variance of the structural displacement is minimized by
theory is applied, Den Hartog [1], the maximum absolute minimizing the area under the absolute value of the frequency
value of the frequency response function of the structure is response function. This corresponds to optimizing the increase
minimized. This corresponds to minimizing the response with in the effective damping ratio obtained by the tuned mass
respect to stationary harmonic excitation with the most critical damper. The optimal parameters for the tuned mass damper
frequency. Alternatively, a stochastic stationary excitation can are obtained numerically and by an approximate analytical
be considered, e.g. white noise. A number of solutions for approach, in which the damping force is assumed to be equal
various types of white noise excitation and various to the force of a conventional tuned mass damper at
minimization criteria have been reported by Warburton [2]. resonance. Furthermore, the influence of the structural
The above mentioned classical results assume that the damping on the solution is investigated.
damping element connecting the structure and the secondary
mass exerts a linear viscous damping force. However, many 2. Equations of motion

∗
Tel.: +45 45 25 19 67; fax: +45 45 88 43 25. A schematic model of the tuned mass damper is shown in
E-mail address: [email protected]. Fig. 1. The equations of motion are given by

0141-0296/$ - see front matter c 2006 Published by Elsevier Ltd

doi:10.1016/j.engstruct.2006.01.006
 F. R¨udinger / Engineering Structures 28 (2006) 1774–1779 1775

simple linear oscillator excited by white noise, and the

standard deviation of the response is given by
x
= = c1 k1 (6)
2
0 2ζ1 ω13

π SW m 1 π SW

which is a measure of the magnitude of the displacement. x0 is

used as a length scale of the response. Choosing ω1−1 as a time
scale, the non-dimensional displacements and time are given
by
Fig. 1. Schematic model of a tuned mass damper.
xi
(m1 + m2)x¨1 + m2x¨2 + c1 x˙1 + k1x1 = m1 W (t) x
(1) yi = 0 ,τ = ω1t. (7)
m2 x¨1 + m2x¨2 + Fd + k2x2 = 0. With this non-dimensional time the single-degree-of-freedom
system has a period of 2π . Introducing these non-dimensional
m 1 is the effective mass of the structure or structural component variables the equations of motion can be written as
(primary mass) and m2 is the mass of the absorber (secondary
mass). m2 will often be much smaller than m1 for practical (1 + μ)y¨1 + μy¨2 + 2ζ1 y˙1 + y1 = U (τ )
(8)
reasons. k1 is the effective stiffness of the structure and k2 is the (ν ) 2
stiffness of the device connecting the absorber to the structure. y¨1 + y¨2 + ηy2 + γ y2 = 0
The structure is assumed to have a small linear viscous damping (ν )
where y2 is the νth derivative of y2 with respect to τ . μ, γ and
with damping coefficient c1. The displacement of the structure is η are given by
denoted x1 and the relative displacement between the structure μ=
m1 , γ =
ω1 , ω2 = m2 , η =
m2 ω12−ν
.(9)

m ω
and the vibration absorber is denoted x2. A dot indicates the 2 2 k2 β
derivative with respect to time. The damping
device connecting the primary and secondary mass exerts a
force Fd given by μ is thus the mass ratio and γ is the ratio between the
ν frequency of the damper ω2 and the frequency of the
d x2 structure. η is a non-dimensional coefficient for the
Fd = β damping force. The intensity of the non-dimensional white
dtν (2) noise U (τ ) can be shown to be
and is thus proportional to the νth derivative of the relative 2ζ1
displacement. For ν = 1 the force is a linear viscous damping
force and for ν = 0 the force is a linear restoring force. It is SU = π . (10)
assumed that 0 ≤ ν ≤ 1 and in the interval 0 < ν < 1 the device The non-dimensional system is seen to depend only on the
contributes with both stiffness and damping. β is the coefficient
and has the dimension mass·time ν−2. For 0 < ν < 1 the νth parameters μ, ζ1, ν, η and γ . The first three parameters are
derivative of x2(t) is defined by the following operator assumed to be given, and the last two are assumed to be the
ν t design variables. It is observed that the standard deviation of
d x 1 d x2(τ ) the structure is unity in this formulation, when the absorber is
2

ν ν absent.
dt =Γ (1 − ν) dt −∞ (t − τ ) dτ (3)
i.e. a combination of a differential and an integral operator, 3. Response magnitude
see e.g. Bagley and Torvik [3]. m1W (t) is the external force,
which is assumed to be a zero-mean white noise, In the frequency domain the excitation and response are
related through the frequency response function. The
E[W (t)W (t + t)] = 2π SW δ( t) (4) frequency response function can be determined by assuming
where scaling by m1 is chosen for convenience. E[ ] is the harmonic excitation and harmonic response:
mean value operator and δ( ) is the Dirac delta function. SW is irτ irτ
the intensity of the white noise W (t). yj=Aje ,U (τ ) = U0e . (11)
Initially, the primary structure is considered without the The white noise excitation U (τ ) from Eq. (8a) has here
temporarily been replaced by a harmonic force. r is a non-
absorber. This corresponds to the case where m2 = 0. The dimensional excitation frequency and i is the imaginary unit. The
equation of motion is in this case given by frequency response function is defined as the ratio between the
2 structural displacement and the excitation. Substituting the
x¨1 + 2ζ1ω1 x˙1 + ω1 x1 = W (t), expressions (11) into the non-dimensional equations of motion
ζ1 = 2√m1k1 ,ω1 = m1 (5)
(8), the following expression is obtained:
c1 k1
A1
H (r ) =
U0
where ζ1 is the damping ratio and ω1 is the natural frequency for 2
r − η(ir ) − γ
ν 2
the structure without the absorber. The system is seen to be a = 2 2 ν 2 4.
(−(1 + μ)r + 2ζ1ir + 1)(r − η(ir ) − γ ) + μr
(12)
1776 F. R¨udinger / Engineering Structures 28 (2006) 1774–1779

Fig. 2. Optimal frequency ratio as a function of mass ratio, (–) ζ1 = 0, (- -) ζ1 = 0.01, (– · –) ζ1 = 0.03, (· · ·) ζ1 = 0.1, (a) ν = 1, (b) ν = 0.75, (c) ν = 0.5,
(d) ν = 0.25.
The variance of the structural response can be evaluated as is given by, see e.g. Warburton [2] or Ankireddi and Yang [6],
σ S
y1 = U −∞ | H (r )| dr. (13) γ opt = ηopt = 2(1 + μ)2
,
4(1 +
μ)3

2 ∞ 2 2 μ μ(4 3μ)
+ +
The integral in this equation is difficult to evaluate analytically
and in the following it will be evaluated numerically. The for ζ1 = 0, ν = 1. (16)
standard deviation of the response σy1 is a measurement The corresponding optimal effective damping increase is given
of the magnitude of the structural displacement. Another by
measurement of the effect of the tuned mass damper can
be obtained by considering the effective damping of the ζ opt μ(1 + μ) for ζ 0, ν 1. (17)
system. By analogy with a single-degree-of-freedom system the = 16 + 12μ 1= =
effective damping ratio is introduced as, see Chang [5], If ζ1 = 0 or ν = 1 the optimal solution is obtained by the
ζ
π SW 1 MATLAB function FMINSEARCH. This function is used to find
ζe = 2σx21 ω13 = σy21 . (14) the values of γ and η, which give the maximum value of ζ
defined in (15). The integral in this expression is evaluated
The effective damping increase is then defined as numerically.
π In Figs. 2–4 the optimal values of γ , η and ζ are shown
ζ = ζe − ζ1 = ∞ 2 − ζ1 (15) as a function of the mass ratio μ for various combinations of
4 0 | H (r )| dr opt
where the fact that H (r ) is the complex conjugate of H ( r) ζ1 and ν. In Fig. 2(a)–(d) the optimal frequency ratio γ is
− shown. Four different values of the fraction are considered:

and therefore | H (r )| = | H (−r )| has been taken into account. ν 1, 0.75, 0.5, 0.25. In all four subfigures four values of the
In the following section optimal parameters of the tuned mass =
structural damping ratio are considered: ζ1 = 0, 0.01, 0.03, 0.1.
damper are defined as the parameters which optimize the The structural damping is seen to have very little influence on
opt
effective damping ratio ζ . the optimal value γ . It is observed that the optimal frequency
ratio is decreasing with increasing values of the mass ratio and
4. Optimal control with decreasing values of the fraction ν.
In Fig. 3(a)–(d) the optimal value of the damping coefficient
If the structural damping is neglected (ζ1 = 0) and the η is shown for the same parameter combinations as the ones
tuned mass damper has linear viscous damping (ν = 1) an considered in Fig. 2(a)–(d). The structural damping is seen to opt

analytical solution in terms of optimal frequency ratio γ and have practically no influence on the optimal value η (the four
curves in each subfigure cannot be distinguished). The
optimal damping coefficient η can be determined. The solution optimal
 F. R¨udinger / Engineering Structures 28 (2006) 1774–1779 1777

Fig. 3. Optimal damping coefficient as a function of mass ratio, (–) ζ1 = 0, (- -) ζ1 = 0.01, (– · –) ζ1 = 0.03, (· · ·) ζ1 = 0.1, (a) ν = 1, (b) ν = 0.75, (c) ν = 0.5, (d)
ν = 0.25.

Fig. 4. Optimal effective damping increase as a function of mass ratio, (–) ζ1 = 0, (- -) ζ1 = 0.01, (– · –) ζ1 = 0.03, (· · ·) ζ1 = 0.1, (a) ν = 1, (b) ν = 0.75, (c) ν =
0.5, (d) ν = 0.25.
damping coefficient increases with increasing values of μ and damping is obtained for the smallest values of the structural
with decreasing values of ν. damping. However, it should be observed that the optimal
opt
In Fig. 4(a)–(d) the optimal effective damping increase is effective damping ζe defined in (14) is largest for the largest
shown, again for the same parameters as in Figs. 2 and 3. In opt
values of the structural damping. ζ is seen to increase with
this case the structural damping is seen to have a small but increasing values of μ. Comparing the four subfigures in Fig. 4 it
visible effect on the result. The largest increase in the is concluded that the value of ν has insignificant influence on
1778 F. R¨udinger / Engineering Structures 28 (2006) 1774–1779

the effective damping, which can be obtained by the tuned

mass damper.
If ν = 1 an approximate analytical solution in terms of the
optimal values of γ and η can be obtained from the solution
given by (16) for ν = 1 and ζ1 = 0. It is assumed that the
parameter values are close to optimal if the combined stiffness
and damping force is equal to the force of the linear viscous
device (ν = 1) at the non-dimensional resonance frequency
ω/ω1 = 1. In the frequency domain this can be expressed as
cos 2 π ν + i sin 2 π νηopt + (γ )
opt 2

1 1

opt opt 2
iην=1 + (γν=1) for ζ1 = 0 (18)
opt opt
where γ and η are the solutions given in (16). If the real
ν=1 ν=1
and imaginary part of the complex force amplitude are set equal
to one another, the following solution is obtained:
1
sin( π ν)
opt
ν =1 − 2

η 1 opt
2

opt
ηopt ν=1 , γ opt (γ )2 cos πν η

for ζ1 = 0. (19)
For ν → 1 the results in (16) are retrieved. In Fig. 5 the
approximate solution (19) is compared to the optimal solution
opt Fig. 5. (a) Optimal frequency ratio and (b) optimal damping coefficient as a
obtained numerically. In Fig. 5(a) γ is shown as a function of
function of mass ratio for ζ1 = 0, (–) ν = 0.75, (- -) ν = 0.5, (– · –) ν = 0.25,
the mass ratio for ζ1 = 0 and three values of the fraction: ν = (•) Eq. (19).
0.75, 0.5, 0.25. The lines correspond to the solutions obtained
numerically and are thus identical to the lines in Fig. 2(b)–(d) for However, even if it is impractical or too expensive to use
polymeric materials in connection with the installation of a
ζ1 = 0. The dots correspond to the approximate analytical
tuned mass damper, the idea of using fractional derivatives in
solution (19). As seen from the figure the approximation is
opt the description of material behaviour can be used for many
reasonably accurate. In Fig. 5(b) η is shown as a function of types of materials, even though a more complicated
μ for the same parameter combinations. The approximation is constitutive equation may be required. Makris and
in this case extremely accurate. The results in (19) are Constantinou [8] have shown that a fractional Maxwell model
probably sufficiently accurate for practical applications.
can be used to give a very accurate description of elastomeric
Based on the analysis in the previous paragraph the results dampers, which are commercially available and have been
presented in Figs. 2–4 are easy to understand. As the value of ν
(ν ) used as damping elements in structures. It is currently being
decreases a larger part of the force ηy2 will be in phase with investigated if the approximate method described in the
the displacement and therefore contribute to the stiffness. This previous section can be generalized to a tuned mass damper
implies that a larger damping coefficient η is needed to reach with elastomeric damping elements. These types of dampers
the desired damping level (see Fig. 3), and that less stiffness γ have high performance compared to viscous dampers, and
2
is needed to reach the desired stiffness level (see Fig. 2). may therefore be preferred if they are equally effective.
When the damping and stiffness parameters are correctly It is generally difficult to work with constitutive laws
adjusted they supply the same damping and stiffness at involving fractional derivatives in the time domain, since the
resonance as the system with linear viscous damping (ν = 1), integral (3) would need to be reevaluated at every time step.
and the effect in terms of reduction in structural vibration is However, the fact that an optimal viscoelastic tuned mass
therefore the same (see Fig. 4). damper has approximately the same effect as an optimal
viscous tuned mass damper suggests that the former can be
5. Discussion replaced by the latter in the time domain analysis as an
approximation. Assessment of the accuracy of such a
The fractional law of the damping element considered has procedure would of course require further investigation.
been shown to give accurate descriptions of the behaviour of Finally, it should be mentioned, that optimal parameters for a
1
polymeric materials with ν 2 , see e.g. Gaul et al. [7]. To the conventional tuned mass damper are approximately equal for
authors knowledge, such types of materials are not currently white noise and harmonic excitation [1,2]. So even though the
used for tuned mass dampers, but the results in the previous idea of an ideal broad band excitation process is an
section show that such a mechanism can be as effective as a idealization, the optimal parameter values derived in this
conventional tuned mass damper, and may be more cost
paper would probably be close to optimal values for a similar
efficient.
tuned mass damper attached to a structure subjected to
harmonic loading.
 F. R¨udinger / Engineering Structures 28 (2006) 1774–1779 1779

6. Conclusion probably be sufficiently accurate for practical applications.

The analysis shows that tuned mass dampers with fractional
A tuned mass damper, in which the damping element has a derivative damping elements have the same effect as optimal
force proportional to a fractional derivative of the conventional tuned mass dampers, if they are tuned correctly.
displacement has been considered. The structure is assumed to
be excited by white noise, and optimal parameters for the Acknowledgement
tuned mass damper are determined by optimizing the effective
damping of the structure–tuned-mass-damper system. The The present project has been supported by the Danish
solution is obtained by numerical calculations and by an Technical Research Council. The support is gratefully
approximate analytical approach where the total force acknowledged.
between the structure and the secondary mass is chosen such
that it matches the force at resonance of a conventional tuned References
mass damper, for which the solution is known. The effect of
structural damping on the solution is investigated. [1] Den Hartog JP. Mechanical vibrations. New York: McGraw-Hill; 1956.
[2] Warburton GB. Optimum absorber parameters for various combinations
The numerical results show that the optimal stiffness of the of response and excitation parameters. Earthquake Engineering and
mass damper decreases with decreasing values of the fraction of Structural Dynamics 1982;10:381–401.
the fractional derivative. The viscoelastic coefficient shows the [3] Bagley RL, Torvik PJ. A theoretical basis for the application of fractional
opposite behaviour. This is due to the fact that the contribution to calculus to viscoelasticity. Journal of Rheology 1983;27:201–10.
[4] Bagley RL, Torvik PJ. On the fractional calculus model of viscoelastic
the stiffness supplied by the viscoelastic force of the damping
behavior. Journal of Rheology 1986;30:133–55.
element increases with decreasing values of the fraction. The [5] Chang CC. Mass dampers and their optimal designs for building
effective damping ratio obtained by inserting the tuned mass vibration control. Engineering Structures 1999;21:454–63.
damper does not depend on the fraction of the derivative in the [6] Ankireddi S, Yang HTY. Simple ATMD control methodology for tall
expression for the viscoelastic force. The structural damping has buildings subject to wind loads. Journal of Structural Engineering 1996;
122:83–91.
very little influence on the solution for optimal parameters, but it
[7] Gaul L, Klein P, Kemple S. Damping description involving fractional
does have a small influence on the optimal effective damping operators. Mechanical Systems and Signal Processing 1991;5:81–8.
increase. The approximate analytical solution is seen to agree [8] Makris N, Constantinou MC. Fractional-derivative Maxwell model for
very well with the numerical solution and would viscous dampers. Journal of Structural Engineering 1991;117:2708–24.