David A. W.

Barton1
Department of Engineering Mathematics,
Energy Harvesting From
University of Bristol,
Queen’s Building,
Bristol BS8 1TR, UK
Vibrations With a Nonlinear
e-mail: [email protected]
Oscillator
Stephen G. Burrow In this paper, we present a nonlinear electromagnetic energy harvesting device that has a
e-mail: [email protected] broadly resonant response. The nonlinearity is generated by a particular arrangement of
magnets in conjunction with an iron-cored stator. We show the resonant response of the
Lindsay R. Clare system to both pure-tone excitation and narrow-band random excitation. In addition to
e-mail: [email protected] the primary resonance, the superharmonic resonances of the harvester are also investi-
gated and we show that the corresponding mechanical upconversion of the excitation
Department of Aerospace Engineering, frequency may be useful for energy harvesting. The harvester is modeled using a Duffing-
University of Bristol, type equation and the results are compared with the experimental data.
Queen’s Building, 关DOI: 10.1115/1.4000809兴
Bristol BS8 1TR, UK

1 Introduction metric nonlinearities to construct a spring with effectively zero

linear stiffness to enable low-frequency vibration isolation; Mc-
Many existing vibration-powered energy harvesting devices are
Farland et al. 关16兴 exploited the concept of energy pumping 关17兴
designed around the principle of linear resonance 关1–5兴 where an 共unidirectional energy transfer兲 to localize vibrational energy in a
inertial mass mounted on a spring damper is excited at the reso- nonlinear attachment; also see the review article by Ibrahim 关18兴
nant frequency of the spring. The benefits of this arrangement are and references therein.
clear: strong amplification of source vibrations and mechanically In this paper, we investigate the behavior of a nonlinear energy
very simple. However, there are also numerous drawbacks, par- harvester that has a broadband resonant response. The device uses
ticularly when the geometry of the device is constrained in some electromagnetic transduction and uses materials that have a high
way 共for example, when the maximum allowed displacement of magnetic permeability. The use of such materials results in
the inertial mass is smaller than the excitation, as is the case when strongly nonlinear mechanical characteristics, which is why they
harvesting from human motion兲. One of the main limitations of a are typically avoided in electromagnetic energy harvester designs
linear mass-spring-damper based energy harvester is that its reso- 共which mostly use air-cored stators to ensure linear behavior兲. The
nant peak is very narrow 关6,7兴. Thus, if the excitation frequency use of high permeability materials enables far stronger coupling
deviates from the resonant frequency of the harvester, very little between the mechanical and electrical domains and, consequently,
power is generated. There have been several attempts to design better energy extraction. In order to characterize the behavior of
tunable harvesters, either by changing the electrical loading 关4兴 or the harvester in “real-world” situations, we subject it to both pure
by actively changing the mechanical properties of the system sinusoidal excitation and narrow-band Gaussian white noise exci-
关8,9兴. However, these devices are in the early stages of develop- tation with varying electrical loads. In reality, the excitation will
ment. be neither purely deterministic or purely random but, instead,
One area of research that has seen little progress until recently somewhere between these two extremes. We note that, in some
is the design of energy harvesters with nonlinear resonant behav- environments, the spectrum of the excitation may be much wider.
ior. Such devices offer the potential for broadband or multiple The overview of this paper is as follows. In Sec. 2, the design
resonant responses allowing use in a wide variety of different of the nonlinear harvester is presented and discussed. Section 3
environments. Initial work by Mann and Sims 关10兴 has shown details the mathematical modeling of the harvester and its quali-
tative reduction to a Duffing-type equation. The experimental re-
how nonlinear devices are more tolerant to manufacturing imper-
sults are presented in Sec. 4: The effect on the frequency response
fections due to their higher bandwidth 共making tuning less impor-
of the harvester is shown for varying excitation strength and fre-
tant兲. Triplett and Quinn 关11兴 studied a weakly nonlinear model of quency and varying electrical loading. Furthermore, superhar-
a piezoelectric energy harvester and demonstrated that, in certain monic resonances are investigated, which shows that there is a
regimes, the nonlinear effect enhanced the performance of the strong possibility for using nonlinear devices to produce usable
device but, in other regimes, detracted from the performance. Cot- power levels at low frequencies. Conclusions are given in Sec. 5.
tone et al. 关12兴 and Gammaitoni et al. 关13兴 investigated the use of
a tunable monostable/bistable energy harvester in the presence of
noise and have shown that the exploitation of nonlinearity can
improve the harvesting performance. On the electrical side, non-
2 Nonlinear Harvester Design
linear electrical coupling to a piezoelectric patch has been shown
by Guyomar et al. 关14兴 to improve power output significantly. In Figure 1 shows a schematic diagram of the nonlinear energy
the related area of vibration absorption, nonlinear structures have harvester that is investigated in this paper. The harvester consists
been used much more extensively: Carrella et al. 关15兴 used geo- of a mass and a set of neodymium 共NdFeB兲 magnets mounted at
the tip of a cantilever beam in close proximity to a laminated iron
stator. The magnets are arranged with alternating polarities in the
1
Corresponding author. 共x , y兲-plane such that a closed magnetic circuit exists with the
Contributed by the Technical Committee on Vibration and Sound of ASME for magnetic flux passing through the stator 共and coil兲, as illustrated
publication in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 22,
2009; final manuscript received December 2, 2009; published online March 17, 2010. in Fig. 1共b兲. The magnets are also arranged in the 共x , z兲-plane with
Assoc. Editor: Bogdan I. Epureanu. alternating polarities to ensure that the flow of magnetic flux re-

Journal of Vibration and Acoustics Copyright © 2010 by ASME APRIL 2010, Vol. 132 / 021009-1

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 Magnets
Excitation Coil

Iron
z y

x x
(a) (b)

Fig. 1 „Top… A photograph of the harvester. „Bottom… A schematic diagram

of the harvester. The harvester consists of a cantilever beam with a tip
mass. The magnets on the tip are arranged such that there is a complete
reversal of magnetic flux during 1 cycle.

verses direction as the beam vibrates. A voltage is generated machining and placement of the magnetic parts. Consequently,
across the coil in proportion to the rate of change of magnetic zero stiffness can never be achieved in practice but it is possible to
flux. get close; in the experiments carried out on this device, it was
The use of a laminated iron-cored stator provides a high degree possible to reduce the first natural frequency of the beam from
of coupling between the mechanical and electrical domains with- approximately 36 Hz to approximately 17 Hz by changing the size
out the need for a large number of windings on the coil, unlike of the air gap without the aid of high-precision instruments.
air-cored designs. Many electromagnetic energy harvesters in the For the experimental device investigated in this paper, we have
literature use air-cored transducers to eliminate the extra damping focused on a parameter regime in which the device is monostable.
caused by eddy currents and magnetic hysteresis; however, this Operating in the bistable regime does not appear to provide sig-
results in poor coupling since air has a very low magnetic perme- nificant advantages, although the magnetic coupling 共flux linkage兲
ability. Furthermore, the air-cored designs may have greater elec- is stronger. A thorough investigation of the benefits of using a
trical losses in their coils due to the increased coil length needed bistable energy harvester is currently in progress and is beyond
to obtain a usable output voltage. It is not yet clear which type of the scope of the present paper. One significant consequence of the
losses are more significant in energy harvester design. bistable configuration, which we note here, is that the motion of
While the use of iron in the energy harvester gives more opti- the tip mass may be chaotic if the excitation force is not suffi-
mal coupling, the interaction of the magnets with the iron stator ciently strong and this is not desirable for providing useful elec-
also gives rise to large nonlinear effects in the mechanical domain. trical output.
The magnetic forces effectively reduce the mechanical stiffness of
the beam for small displacements. Thus, the overall effect is to
create a hardening spring that has a lower natural frequency than
the beam without the magnets attached. The reduction in the natu- 3 Mathematical Modeling
ral frequency is determined by the strength of the magnets and the The mechanical part of the energy harvester, as shown in Fig. 1,
size of the air gap between the magnets and the iron stator; if the can be modeled as a cantilever beam using the Euler–Bernoulli
air gap is decreased sufficiently, the device becomes bistable and equation with appropriate boundary conditions 关20兴. However, it
the tip mass will hop between the two stable states when given a is convenient to make the simplifying assumption that the beam is
strong enough perturbation. Consequently, this design of energy operating solely in its first mode. This is justified by the fact that
harvester enables the resonant behavior of the device to be tuned the harvester is always excited at or below its first resonant fre-
significantly with only small mechanical changes. quency 共approximately 24 Hz in the experiments below兲 and the
If, when in the bistable regime, the stable states were equally second resonant frequency is far higher 共approximately 160 Hz in
favored 共i.e., they lie in symmetric potential wells兲, it would be the experiments below兲. Consequently, the mechanical part of the
possible to make an energy harvester with almost zero linear stiff- energy harvester is modeled as the mass-spring-damper system
ness by changing the air gap appropriately. This would be ex-
tremely useful for energy harvesting from very low-frequency vi- mü + cmechu̇ + kmechu + Fmag共u,u̇兲 = − mv̈
brations, as well as enabling designs that incorporate more exotic
ideas such as energy pumping 关19兴. Furthermore, in small scale where u is the tip displacement, m is the equivalent mass, cmech is
designs, where it is not always possible to design a linear resona- the 共modal兲 mechanical damping, kmech is the mechanical spring
tor with an appropriate natural frequency due to size constraints, constant, Fmag is the magnetic interaction force, and v is the dis-
this type of design could be successfully employed. However, due placement applied to the base of the energy harvester.
to the extreme sensitivity of the magnetic field to imperfections in The magnetic part of the system is significantly more difficult
the device, it is not possible to do this without high-precision to model due to the distributed nature of the fields involved. Con-

021009-2 / Vol. 132, APRIL 2010 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 1 5 6 vester. The tests were primarily frequency sweeps with either
1 0 4 sinusoidal or narrow-band noise as the excitation. Both the pri-
mary resonance and the superharmonic resonances were consid-
5 2
ered.

V o lta g e (V )
F o rc e (N )
0 0 When used in practice, it is unlikely that the energy harvester
will experience either of the two extremes of excitation 共that is,
5 2
purely sinusoidal or purely random兲 and the “real” excitation will
1 0 4 lie somewhere in the middle. For example, when harvesting en-
(a ) (b ) ergy from machine vibrations, the excitation is likely to be well
1 5 6
4 3 2 1 0 1 2 3 4 - 0 .8 - 0 .4 0 0 .4 0 .8 defined but have a significantly more complicated spectrum than
D is p la c e m e n t (m m ) V e lo c ity (m s 1 ) just a simple sinusoid. Whereas when harvesting from motion of
people or animals, the excitation will be much more random but
Fig. 2 „a… Mechanical spring characteristics of the beam/ still have some well defined characteristics. However, without a
magnet system. The solid gray curve and the dashed gray line particular application in mind, these two extremes provide a guide
show the magnetic and mechanical spring characteristics,
respectively; the solid black curve shows the overall spring
to how the harvester will behave in reality.
characteristics. „b… Characteristics of the electromagnetic 4.1 Experimental Setup. Base excitation of the energy har-
induction. vester was achieved using an electrodynamic shaker with a
closed-loop position feedback controller to ensure reliable opera-
sequently, as a first approximation, we assume the magnetic forces tion across all frequency ranges. Additionally, the shaker was ori-
to be modeled well by a cubic polynomial in displacement and a entated horizontally to ensure that no extra asymmetries were in-
linear function with respect to velocity: troduced into the system by the action of gravity. The energy
harvester was instrumented with PCB Piezotronics accelerometers
Fmag ª kmagu + ␤magu3 + cmagu̇ + Felec on its base and on the tip mass to provide positional measure-
where Felec is the force reflected back from the electrical domain. ments in addition to the voltage output of the coil.
The velocity term acts as an effective magnetic damping incorpo- When the harvester was excited with sinusoidal forcing, the
rating any losses due to the magnetic field’s interaction with the position and velocity information were determined by numerically
iron-cored stator 共e.g., eddy current losses and magnetic hyster- integrating the accelerometer output and determining the integra-
esis兲. The coefficients of the polynomials are determined by fitting tion constants by the periodicity. Since the forcing frequency was
to experimental data, see Fig. 2. While this might seem a gross relatively high, the drift of the accelerometer signal across one
simplification of the physics, it provides a good approximation of oscillation was negligible and so the position and velocity infor-
the behavior of the energy harvester, as shown by the experimen- mation obtained was reasonably accurate.
tal data. Furthermore, the alternative approaches, such as finite When the harvester was excited with narrow-band random forc-
element electromagnetics, are very cumbersome and very few, if ing 共and so the response was no longer periodic兲, the accelerom-
any, generally applicable design rules can be derived from them. eter data was first filtered to remove low-frequency components
The bidirectional coupling between the magnetic and electrical 共frequencies below 0.1 Hz兲 before numerical integration. The re-
domains occurs since a changing magnetic field strength 共due to sulting positional information had approximately a zero-mean
the change in position of the permanent magnets兲 will induce a suggesting that this was sufficient to remove the effects of accel-
potential difference across the coil. In turn, this potential differ- erometer drift.
ence will drive the flow of current around the electrical circuit and The narrow-band random signal was created by passing Gauss-
the flowing current will create a magnetic field in opposition to ian white noise through a linear band-pass filter of the form
the original changes in the magnetic field.
f̈ + ␥ ḟ + ␻2f f = ␥1/2␻ f W 共2兲
The relationship between the position of the magnets/tip mass
and the magnetic field strength in the stator is approximately lin- where ␻ f is the center frequency, ␥ is the bandwidth of the filter,
ear during normal operation, as shown by Fig. 2. Consequently, and W is a Gaussian white noise source 关23兴. This was imple-
there is a one-to-one relationship between the velocity of the mag- mented in a dSpace real-time controller, which also enabled auto-
nets and the potential difference induced across the coil. mated testing of the harvester.
The final nonlinear model is derived by applying Kirchhoff’s
circuit equations to the simple electrical load of two resistors 共one 4.2 Physical Characteristics. The system parameters of Eq.
for the coil resistance and one for the load resistance兲 and an ideal 共1兲 were determined by performing an initial frequency sweep
voltage source in series. This gives with the coil in an open-circuit state. At a discrete set of frequen-
cies the periodic motion of the harvester was recorded. By com-
mü + 共cmech + cmag兲u̇ + 共kmech + kmag兲u + ␤magu3 = − mv̈ − ␪i bining all these measurements, it was possible to reconstruct the
phase space of Eq. 共1兲 as a 共position, velocity, and acceleration兲
␪u̇ = Ri triplet and so measure the spring and damping characteristics; in
where ␪ is a coupling coefficient and R ª Rcoil + Rload is the total the systems identification literature, this is known as the restoring
resistance in the electrical circuit. These two equations can be force surface method 关24兴.
combined to give the Duffing-type equation Figure 2共a兲 shows the reconstructed force-displacement charac-
teristic of the harvester for zero velocity. Over the region of op-
mü + 共cmech + cmag + ␪2/R兲u̇ + 共kmech + kmag兲u + ␤magu3 = − mv̈ eration shown, the force-displacement characteristic has a strong
共1兲 cubic element with minimal asymmetry. Consequently, the dy-
namics of the harvester are similar to that of a hardening spring.
The Duffing equation has been extensively studied in a wide Figure 2共b兲 shows the reconstructed voltage-velocity characteris-
variety of contexts, and methods of solution can be found in any tic of the harvester at peak displacement. This validates the as-
good textbook on nonlinear ordinary differential equations 共see, sumption of a linear relationship between the velocity and voltage,
for example, Refs. 关21,22兴兲. For brevity, we omit the details here. which only starts to break at higher velocities.
The measured parameters for the model Eq. 共1兲 are as
4 Experimental Results follows: m = 80 g, cmech = 0.06 N s m−1, cmag = 0.24 N s m−1,
This section describes the experimental rig and the tests under- ␪ = 7 V s m−1, 12 ⍀ ⱕ R ⱕ 212 ⍀, kmech = 4.09⫻ 103 N m−1,
taken to investigate the dynamics of the nonlinear energy har- kmag = −2.37⫻ 103 N m−1, and ␤mag = 1.63⫻ 108 N m−3.

Journal of Vibration and Acoustics APRIL 2010, Vol. 132 / 021009-3

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 0 .8 response. For large amplitude excitations, the resonant behavior of
0 .1 5 m m e x c ita tio n
0 .1 0 m m e x c ita tio n
0 .7 0 .0 5 m m e x c ita tio n the system is greatly increased and covers a wide band of frequen-
cies, resulting in a large hysteresis loop where two stable states
0 .6
1) coexist 共one high-energy state and one low-energy state兲. Thus, at
0 .5 worst the harvester appears linear and the advantages of the non-
P e a k v e lo c ity (m s

0 .4 linearity are not present but at best there is a dramatic increase in

the useful bandwidth.
0 .3
Figure 4 presents a comparison of the experimental data and the
0 .2 model, which shows reasonable agreement between experiment
0 .1
and theory. The response of the model was computed using the
numerical continuation software AUTO-97 关25兴 rather than the ana-
0 .0
1 0 1 5 2 0 2 5 3 0 3 5 4 0 lytical formulas in Refs. 关21,22兴 due to the significant third har-
F re q u e n c y (H z) monic component found in the response. The source of the dis-
crepancies around the point of linear resonance are as yet
Fig. 3 Experimental frequency sweeps „increasing frequency unknown, but are likely to be due to the complicated damping
then decreasing frequency… for three amplitudes of excitation. mechanisms of magnetic hysteresis and eddy currents and the
The peak excitation displacement is kept constant across a fre-
presence of small asymmetries in the harvester.
quency sweep using a closed-loop controller around the
shaker. It is not surprising that the top branch of experimental results
does not extend as far as the theory predicts. As the frequency of
excitation is increased, the high-energy state becomes only weakly
4.3 Primary Resonance: Sinusoidal Forcing. Figure 3 stable, its basin of attraction shrinks, and, consequently, small
shows the fundamental behavior of the harvester as the excitation disturbances cause the harvester to be knocked into the low-
frequency varies. The velocity of the tip mass is plotted against energy state. With this in mind, we make the general comment
frequency since this is the most important factor when considering that the resonant peaks determined by nonlinear analysis of, for
the power output of the harvester. The forcing is a sinusoid with a example, the Duffing equation should be treated with great care
fixed amplitude of displacement and the coil is in an open-circuit since the solution may only have a small basin of attraction, which
state. Data are shown for three different forcing amplitudes: is not useful from a practical view point. However, it is relatively
0.1 mm, 0.2 mm, and 0.3 mm peak to peak displacement. The straightforward to determine the approximate size of the basin of
frequency is increased in steps of 0.2 Hz from 10 Hz to 40 Hz and attraction by numerical simulation.
then decreased back to 10 Hz to show the hysteresis in the It is relevant to note that in general nonlinear systems, there
response. may exist isolated solutions 共isolas兲 in the frequency response that
For small amplitude excitations, the harvester operates in a are not connected to the primary branch of solutions. However, in
near-linear regime. There is a pronounced increase in the velocity this particular study, no evidence for the existence of isolas was
near the linear resonance of 23.6 Hz and little hysteresis in the found.
Figure 5 shows the basins of attraction of the high-energy state
共white兲 and the low-energy 共black兲 for the model Eq. 共1兲 when 共a兲
5
M o d e l d a ta ␻ = 27.5 Hz, 共b兲 ␻ = 30 Hz, 共c兲 ␻ = 32.5 Hz, and 共d兲 ␻ = 35 Hz.
E x p e rim e n ta l d a ta
The basins are shown relative to the start of a period of forcing
4 共i.e., the instantaneous forcing from the shaker is zero兲. It is im-
mediately clear that the higher frequency, the more dominant the
P e a k d is p la c e m e n t (m m )

3 low-energy state becomes.

Figure 6共a兲 shows the peak velocity against frequency for six
different electrical loading conditions. The electrical load is a vari-
2
able resistor set to one of 50 ⍀, 75 ⍀, 100 ⍀, 150 ⍀, 200 ⍀,
or open-circuit conditions. The experimental results show that
1 there is little change in the peak velocity of the tip mass as the
electrical damping is increased 共equivalently, the resistive load is
0 decreased兲. The primary difference in behavior comes from the
1 0 1 5 2 0 2 5 3 0 3 5 4 0
F re q u e n c y (H z) change in drop down frequency. However, this jump up frequency
remains constant with respect to the electrical loading. It should
Fig. 4 A comparison of the experimental frequency response be noted that the voltage seen across the coil is not invariant with
and the model equation „1… frequency response. The discrep- respect to the electrical damping since the relationship between
ancies are likely to be due to unmodeled magnetic loss velocity and voltage is also dependent on the electrical load.
mechanisms.

(a ) (b ) (c ) (d )
8 8 8 8

4 4 4 4
)
1
V e lo c ity (m s

0 0 0 0

-4 -4 -4 -4

-8 -8 -8 -8
-4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4 -4 -2 0 2 4
D is p la c e m e n t (m m ) D is p la c e m e n t (m m ) D is p la c e m e n t (m m ) D is p la c e m e n t (m m )

Fig. 5 The basins of attraction of the model equation „1… computed by numerical simu-
lation for „a… 27.5 Hz, „b… 30 Hz, „c… 32.5 Hz, and „d… 35 Hz. The basin of attraction of the
low-energy state is colored in black and the basin of attraction of the high-energy state
is colored in white.

021009-4 / Vol. 132, APRIL 2010 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 0 .8 0 .1 0
O p e n c irc u it (a ) (b )
0 .7 2 0 0 lo a d
1 5 0 lo a d
1 0 0 lo a d
0 .0 8
0 .6
7 5 lo a d

)
5 0 lo a d

1
0 .5

P e a k p o w e r (W )
P e a k v e lo c ity (m s
0 .0 6
0 .4

0 .3 0 .0 4

0 .2
0 .0 2
0 .1

0 .0 0 .0 0
2 0 2 2 2 4 2 6 2 8 3 0 3 2 3 4 3 6 2 0 2 2 2 4 2 6 2 8 3 0 3 2 3 4 3 6
F re q u e n c y (H z) F re q u e n c y (H z)

Fig. 6 A comparison of frequency sweeps for varying electrical loads: „a…

velocity and „b… power. Note that in the open-circuit condition, no power is
developed.

Figure 6共b兲 shows the peak electrical power developed in the tween the periodic and random excitation is due to the hopping
load against frequency for same experiments. 共Note that the open- between the different energy states as the random excitation
circuit condition results in zero power developed.兲 The optimum changes in strength. For higher frequencies, once the high-energy
peak power condition occurs at approximately 100 ⍀, which is state has been left the random excitation is not sufficiently strong
the same as the resistance required to achieve optimum peak to force the motion back into the high-energy state.
power for a linear harvester with the same damping coefficients. The effect of varying the bandwidth of the random excitation is
When the resistance is decreased further, the peak power drops shown in Fig. 8. The panels show the frequency responses for 共a兲
off. However, the power output at lower frequencies increases.
2 Hz bandwidth, 共b兲 1 Hz bandwidth, 共c兲 0.5 Hz bandwidth, and
Thus, at the edge of the bistable region 共approximately 25 Hz兲, the
共d兲 0.25 Hz bandwidth. It could be argued that the jump up/jump
optimum peak power output is obtained for 50 ⍀.
These results provide further support for the notion that nonlin- down points in the frequency response are most clear in the nar-
ear energy harvesters can have wide-band responses. In effect, the rowest bandwidth case, with the higher bandwidth cases smooth-
bandwidth of the device can be increased considerably by includ- ing out the distinction, but this is a tenuous conclusion based on
ing a variable load resistance into the electrical circuit design. the results shown here. Since there is not a great deal of difference
between the results, this indicates that there is a very sharp tran-
4.4 Primary Resonance: Random Forcing. Figure 7 shows sition at low bandwidth between the frequency response of the
a comparison of the frequency response of the harvester in open- periodic excitation and the frequency responses shown here.
circuit configuration with periodic excitation and narrow-band Further experiments 共not shown兲 when varying the electrical
random excitation 共bandwidth of 2 Hz兲. The vertical axis shows
loading indicate that the peak power output is obtained for very
the average tip velocity of the harvester, which is defined as the
root-mean-square 共rms兲 velocity averaged over 10 s of excitation. low electrical resistances 共approximately 50 ⍀兲, which is consis-
Over the 10 s sampling window, the average rms of the shaker tent with the data presented in Fig. 6 since much of the high-
displacement is approximately equal for both the periodic excita- energy branch is not present.
tion and the random excitation. Overall, the experimental results presented here are in broad
The results in Fig. 7 show that the peak velocity attained by the agreement with theoretical findings on the Duffing equation with
harvester under periodic excitation is never reached by the har- narrow-band excitation 关23,26兴. Simply stated, increasing the
vester under random excitation. However, the harvester motion bandwidth of the narrow-band excitation results in the device ap-
does follow the high-energy branch briefly. The difference be- pearing increasingly linear. If the bandwidth of the random exci-

0 .6
P e rio d ic e x c ita tio n
R a n d o m e x c ita tio n
0 .5
)
1

0 .4
A v e ra g e v e lo c ity (m s

0 .3

0 .2

0 .1

0 .0
1 5 2 0 2 5 3 0 3 5
C e n te r fre q u e n c y (H z )

Fig. 7 A comparison of the frequency response to a pure-tone excitation

and a narrow-band random signal „2 Hz bandwidth… of the energy harvester
in open-circuit conditions. The response of the harvester to the random
signal is averaged „rms… over 10 s.

Journal of Vibration and Acoustics APRIL 2010, Vol. 132 / 021009-5

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 0 .3 0 0 .3 0
(a ) (b )

)
0 .2 5 0 .2 5

1
A v e ra g e v e lo c ity (m s

A v e ra g e v e lo c ity (m s
0 .2 0 0 .2 0
0 .1 5 0 .1 5
0 .1 0 0 .1 0
0 .0 5 0 .0 5
0 .0 0 0 .0 0
1 5 2 0 2 5 3 0 3 5 1 5 2 0 2 5 3 0 3 5
C e n te r fre q u e n c y (H z ) C e n te r fre q u e n c y (H z )
0 .3 0 0 .3 0
(c ) (d )
)

)
0 .2 5 0 .2 5
1

1
A v e ra g e v e lo c ity (m s

A v e ra g e v e lo c ity (m s
0 .2 0 0 .2 0
0 .1 5 0 .1 5
0 .1 0 0 .1 0
0 .0 5 0 .0 5
0 .0 0 0 .0 0
1 5 2 0 2 5 3 0 3 5 1 5 2 0 2 5 3 0 3 5
C e n te r fre q u e n c y (H z ) C e n te r fre q u e n c y (H z )

Fig. 8 Frequency sweeps for varying bandwidth narrow-band random excitation. The corresponding
bandwidths are „a… 2 Hz, „b… 1 Hz, „c… 0.5 Hz, and „d… 0.25 Hz.

tation is decreased sufficiently 共approximately 0.1 Hz兲, most of resonances and 1:2 resonances, respectively. These second two
the hysteresis loop seen in the pure-tone excitation case reappears. resonances are much smaller since they are determined by qua-
dratic characteristics within the harvester 共asymmetries in the
4.5 Superharmonics. Nonlinear devices often exhibit sub/
force-displacement characteristics兲, which are very small.
superharmonic resonances, where the frequency of the response is
lower/higher than the excitation frequency, respectively. Of inter- The superharmonic resonances appear very similar to linear
est here are the superharmonic responses, where the harvester resonances in that they possess no hysteretic regions in the fre-
responses at a higher frequency than the excitation frequency—a quency response. However, in general the superharmonic reso-
mechanical upconversion of the frequency. nances may also take the same form as the primary resonance if
Figure 9 shows the low-frequency response of the nonlinear the nonlinear characteristics are suitably strong.
harvester for a variety of electrical loading conditions. Superhar- Although the superharmonic response is far less than the pri-
monic resonances at approximately 4.6 Hz and 7.8 Hz are clearly mary response, the superharmonic resonances may be useful in
seen. The harvester is responding at its natural frequency when designing very low-frequency energy harvesters or very small en-
excited with a frequency of five times and three times lower, ergy harvesters where it is not possible to build a linear resonator
respectively. These resonances are caused by the dominant cubic of sufficiently low frequency. In addition, the upregulation of the
characteristic of the harvester. 共Note that the 1:5 resonance does frequency means that the overall displacement remains low while
not require a quintic term in the model; it can be due to the cubic providing a relatively high velocity 共compared with the excitation
term.兲 It is also possible to make out two additional superhar- force兲. Consequently, future designs might seek to strengthen
monic resonances at 5.8 Hz and 11.7 Hz; these correspond to 1:4 these superharmonic resonances.

0 .1 0 [1 :3 ]
O p e n c irc u it
2 0 0 lo a d
1 5 0 lo a d
0 .0 8 1 0 0 lo a d
7 5 lo a d
)

5 0 lo a d
1
P e a k v e lo c ity (m s

0 .0 6

[1 :5 ]
0 .0 4

0 .0 2 [1 :2 ]
[1 :4 ]

0 .0 0
4 6 8 1 0 1 2 1 4
F re q u e n c y (H z)

Fig. 9 Superharmonic resonances of the energy harvester for varying elec-

trical loads. At the resonance peaks, the harvester is responding at an inte-
ger multiple of the excitation frequency „as denoted above each peak in the
figure…. At the most prominent peak, the harvester is responding at three
times the excitation frequency.

021009-6 / Vol. 132, APRIL 2010 Transactions of the ASME

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms

 5 Discussion and Conclusions Networks,” Proceedings of the Third International Energy Conversion Engi-
neering Conference, San Francisco, CA.
In this paper, we have shown that a nonlinear energy harvester 关4兴 Renno, J., Daqaq, M., and Inman, D., 2009, “On the Optimal Energy Harvest-
is able to overcome some of the inherent limitations of a linear ing From a Vibration Source,” J. Sound Vib., 320, pp. 386–405.
关5兴 Stephen, N., 2006, “On Energy Harvesting From Ambient Vibration,” J. Sound
energy harvester, namely, that of having a narrow resonant re- Vib., 293共1–2兲, pp. 409–425.
sponse. Moreover, the bandwidth of the harvester can be increased 关6兴 Burrow, S., and Clare, L., 2007, “A Resonant Generator With Non-Linear
further with a varying resistive load. However, the advantages of Compliance for Energy Harvesting in High Vibrational Environments,” Pro-
the wide bandwidth are only apparent when there is a consistent ceedings of the Electric Machines & Drives Conference, IEMDC ’07, Antalya,
Turkey, Vol. 1, pp. 715–720.
共uninterrupted/nonrandom兲 vibration source 共for example, when 关7兴 Burrow, S., Clare, L., Carrella, A., and Barton, D., 2008, “Vibration Energy
the harvester is attached to a piece of rotating machinery兲 due to Harvesters With Non-Linear Compliance,” Proceedings of the SPIE Smart
the coexistence of a low-energy state and a high-energy state. In Structures/NDE Conference, San Diego, CA, p. 692807.
particular, random excitations appear to average out the high- 关8兴 Challa, V., Prasad, M., Shi, Y., and Fisher, F., 2008, “A Vibration Energy
Harvesting Device With Bidirectional Resonance Frequency Tunability,”
energy and low-energy states. We are currently investigating pos- Smart Mater. Struct., 17共1兲, p. 015035.
sible ways of incorporating a restarting mechanism to overcome 关9兴 Morris, D., Youngsman, J., Anderson, M., and Bahr, D., 2008, “A Resonant
sudden drops in the source vibration as found with random exci- Frequency Tunable, Extensional Mode Piezoelectric Vibration Harvesting
tation, which uses a small amount of the previously harvested Mechanism,” Smart Mater. Struct., 17共6兲, p. 065021.
energy to perturb the harvester back into a high-energy state. 关10兴 Mann, B., and Sims, N., 2009, “Energy Harvesting From the Nonlinear Oscil-
lations of Magnetic Levitation,” J. Sound Vib., 319共1–2兲, pp. 515–530.
While the model presented in this paper shows reasonable 关11兴 Triplett, A., and Quinn, D., 2009, “The Effect of Nonlinear Piezoelectric Cou-
quantitative agreement with the experimental results, a much pling on Vibration-Based Energy Harvesting,” J. Intell. Mater. Syst. Struct.,
more in depth analysis of magnetic loss mechanisms is needed to 20, pp. 1959–1967.
fully model the system. A better model of the magnetic losses 关12兴 Cottone, F., Vocca, H., and Gammaitoni, L., 2009, “Nonlinear Energy Harvest-
ing,” Phys. Rev. Lett., 102共8兲, p. 080601.
would also help determine the usefulness of iron-cored stator de- 关13兴 Gammaitoni, L., Neri, I., and Vocca, H., 2009, “Nonlinear Oscillators for
signs. The high magnetic permeability of iron gives a high degree Vibration Energy Harvesting,” Appl. Phys. Lett., 94共16兲, p. 164102.
of coupling between mechanical and electrical domains but it is 关14兴 Guyomar, D., Badel, A., Lefeuvre, E., and Richard, C., 2005, “Toward Energy
still not clear if this benefit outweighs the corresponding magnetic Harvesting Using Active Materials and Conversion Improvement by Nonlinear
Processing,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, 52共4兲, pp. 584–
losses. 595.
One important benefit of using a nonlinear harvester is the pres- 关15兴 Carrella, A., Brennan, M., and Waters, T., 2007, “Static Analysis of a Passive
ence of superharmonic resonances at frequencies well below the Vibration Isolator With Quasi-Zero-Stiffness Characteristic,” J. Sound Vib.,
共linear兲 natural frequency. At these resonances, the harvester re- 301共3–5兲, pp. 678–689.
关16兴 McFarland, D. M., Bergman, L. A., and Vakakis, A. F., 2005, “Experimental
sponds at a higher frequency than the excitation frequency—a Study of Non-Linear Energy Pumping Occurring at a Single Fast Frequency,”
mechanical upconversion of frequency. These superharmonic Int. J. Non-linear Mech., 40共6兲, pp. 891–899.
resonances may enable the design of devices that resonate at low 关17兴 Vakakis, A. F., and Gendelman, O., 2001, “Energy Pumping in Nonlinear
frequencies while still being able to extract moderate amounts of Mechanical Oscillators: Part II—Resonance Capture,” ASME J. Appl. Mech.,
68共1兲, pp. 42–48.
power; this could be useful for applications where it is not pos- 关18兴 Ibrahim, R., 2008, “Recent Advances in Nonlinear Passive Vibration Isola-
sible to build a linear resonator of sufficiently low frequency. tors,” J. Sound Vib., 314共3–5兲, pp. 371–452.
关19兴 Quinn, D., Vakakis, A., and Bergman, L., 2007, “Vibration-Based Energy Har-
Acknowledgment vesting With Essential Nonlinearities,” ASME Paper No. DETC2007-35457.
关20兴 Inman, D., 2007, Engineering Vibration, Prentice-Hall, Englewood Cliffs, NJ.
D.A.W.B. gratefully acknowledges the support of Great West- 关21兴 Jordan, D., and Smith, P., 1999, Nonlinear Ordinary Differential Equations,
ern Research. S.G.B. and L.R.C. gratefully acknowledge the sup- Oxford University Press, New York.
关22兴 Nayfeh, A., and Mook, D., 1995, Nonlinear Oscillations, Wiley, New York.
port of EPSRC Grant No. EP/E044220/1 共Energy Harvesting: Vi- 关23兴 Rajan, S., and Davies, H., 1988, “Multiple Time Scaling of the Response of a
bration Powered Generators With Nonlinear Compliance兲. Duffing Oscillator to Narrow-Band Random Excitation,” J. Sound Vib.,
123共3兲, pp. 497–506.
关24兴 Kerschen, G., Worden, K., Vakakis, A., and Golinval, J.-C., 2006, “Past,
References Present and Future of Nonlinear System Identification in Structural Dynam-
关1兴 Williams, C., and Yates, R., 1996, “Analysis of a Micro-Electric Generator for ics,” Mech. Syst. Signal Process., 20共3兲, pp. 505–592.
Microsystems,” Sens. Actuators A, 52共1–3兲, pp. 8–11. 关25兴 Doedel, E., Champneys, A., Fairgrieve, T., Kuznetsov, Y., Sandstede, B., and
关2兴 Williams, M., and Blakeborough, A., 2001, “Laboratory Testing of Structures Wang, X., 1998, AUTO-97: Continuation and Bifurcation Software for Ordi-
Under Dynamic Loads: An Introductory Review,” Philos. Trans. R. Soc. Lon- nary Differential Equations.
don, Ser. A, 359共1786兲, pp. 1651–1669. 关26兴 Davies, H., and Rajan, S., 1988, “Random Superharmonic and Subharmonic
关3兴 Baker, J., Roundy, S., and Wright, P., 2005, “Alternative Geometries for In- Response: Multiple Time Scaling of a Duffing Oscillator,” J. Sound Vib.,
creasing Power Density in Vibration Energy Scavenging for Wireless Sensor 126共2兲, pp. 195–208.

Journal of Vibration and Acoustics APRIL 2010, Vol. 132 / 021009-7

Downloaded From: http://vibrationacoustics.asmedigitalcollection.asme.org/ on 02/25/2015 Terms of Use: http://asme.org/terms