Modeling Aspects of Nonlinear Energy Harvesting for Increased Bandwidth

Over the last years, the field of energy harvesting has become a promising technique as power supply of autonomous electronic devices. Those use the surrounding energy, such as vibrations, temperature gradients or radiation, for conversion in electrical energy. Mechanical vibrations are an attractive source due to their high availability in technical environments, thus numerous research groups are working on this topic. The most important conversion methods for ambient vibrations are electromagnetic, electrostatic and piezoelectric. All those techniques have been successfully demonstrated in the past. [11] provides an overview of the basics in energy harvesting.


Introduction
Over the last years, the field of energy harvesting has become a promising technique as power supply of autonomous electronic devices.Those use the surrounding energy, such as vibrations, temperature gradients or radiation, for conversion in electrical energy.Mechanical vibrations are an attractive source due to their high availability in technical environments, thus numerous research groups are working on this topic.The most important conversion methods for ambient vibrations are electromagnetic, electrostatic and piezoelectric.All those techniques have been successfully demonstrated in the past.[11] provides an overview of the basics in energy harvesting.
Due to the fact, that vibration energy harvester generates the most energy when the generator is excited at its resonance frequency, the converter needs to be tuned to the main external frequency of the individual environment.If the excitation frequency shifts, the performance of the generator may reduced drastically.In practical use the vibration of an environment may vary in a large spectrum.To overcome this disadvantage researchers work hard to increase the working bandwidth of an energy harvester.
This chapter is a contribution to the current state of the art for modeling broadband energy harvesting generators.In the first part the electromechanical model is derived in terms of using lumped parameters.The system is based on a piezoelectric bimorph structure.The coupled differential equations for the case of a simple electrical circuit are derived and furthermore the possibility to enhance the energy extraction is analyzed.The use of generator arrays to archive a high power outputs in a wide frequency range is discussed.In detail, the Synchronized Switch Harvesting on Inductor (SSHI) technique is studied.Further the modeling of the promising piezomagnetoelastic energy harvesting technique is covered in the last part.

Modeling of piezomechanical structures
Fundamental for the energy harvesting techniques presented in the following is the piezoelectric bimorph.In the following, the general modeling of piezomechanical structures is given, and further on applied to the case of a bending bimorph.
The calculations are based on the potential energy stored in a piezoelement, ( 1 )   where T, S, D, E represent the mechanical stress and strain as well as the electrical displacement and field.V is the volume of the piezoelement.A one-dimensional strain distribution within the piezoelement in axis direction i is assumed.According to [6] the axis of polarization is defined as x 3 .Therefore the transversal effect is represented by i = 1, where the mechanical strain is normal to the direction of polarization, and the longitudinal effect by i = 3, where the mechanical strain is in the direction of polarization.After some mathematical calculations, see [9] the energy can be written as The electrical charge at the electrodes is termed Q p , while the mechanical compliance in axis direction i is given as s ii .Further on, the material coupling of the piezoelement is given by k 3i = d 2 3i / s E ii ε T 33 .In the following, the stiffness c p in x i direction, the capacitance C p of the piezoceramics and the piezoelectric force factor α are introduced as They depend on the area of electrodes A el , the piezoelectric constant d 3i , the permittivity ε 33 as well as the geometry of the piezoelement (length ℓ 3 between electrodes and length ℓ i of the piezoelement in direction of mechanical strain).
The energy terms in Equation 2 can be classified as the stored electrical energy, the stored mechanical energy and the converted energy.For convenience of the following calculations, the mechanical strain is split into the mean value Si ,t h em e a ns t r a i n Si,3 along the x 3 axis (between the electrodes), and the difference ΔS i,3 between the actual strain and Si,3 , The reason for this representation is that piezoelectric systems with a homogeneous strain distribution is readliy described by ΔS i,3 = 0, which strongly simplifies the calculations.Additionally, the influence of an uneven strain distribution can be seen in the term ΔS i,3 .
See Figure 1 for an illustration of these definitions.
In case of a continuous system it is reasonable to discretize it for the further analysis.The mechanical deformation is then described by n degrees of freedom (DOF) q i , while the charge Q p is the electrical DOF, Each mechanical DOF is associated with a global mode shape, which defines the mechanical strain distribution S i within the piezoelectric volume (and the rest of the mechanical system).
The overall strain distribution is then the sum of all mode shapes.In order to rewrite the energy term in Equation 2, the term −ℓ i Si , which represents the mean deformation of the piezoelement in x i direction, will be represented by Here, a mechanical coupling vector κ is introduced.In that form, the energy terms in Equation 2 can be rewritten as The potential energy can be rewritten as ( 8 ) so that the stiffness matrix of the system follows as The term C mech represents the 'mechanical' stiffness matrix of the piezoelement, which can be deduced in the same way as standard mechanical systems.

Piezoelectric bimorph
Now we can apply the above obtained results for the piezoelectric bimorph.We are considering the general case of a piezoelectric layer which has a distance e to the neutral axis of the beam.The coordinate axes are defined in such a way that the origin is at contact between the piezoelectric and the substrate layers at the clamped end, see Figure 2. The x 1 axis is in beam direction and the deformations occur in x 3 direction, which is also the direction of polaziration.With Euler-Bernoulli assumptions the strain is only applied in x 1 direction.That means the transversal effect of the piezoceramics is utilized.The strain terms described in the previous section then read for the case of the clamped beam: The bending of the beam is described by w(x 1 , t).This term will be split into the part depending on coordinate x 1 and the part depending on time t, w(x 1 , t)=W(x 1 )q(t), w ′′ (x 1 , t)=W ′′ (x 1 )q(t).
In this example, only one mechanical degree of freedom is used to describe the vibrations.This is typically a reasonable approximation when the system vibrates close to one of its eigenfrequencies.In this way, the general stiffness matrix according to Equation 9reduces to and the mechanical coupling is written as while the piezoelectric coupling reads Terms of the kind α 2 C p κ 2 , which determine the increase in eigenfrequencies between short circuit electrodes and isolated electrodes are obtained as For a better understanding of these terms, it is useful to introduce the area moment of inertia I PZT of the piezoceramics around its own center of gravity and the moment of inertia I nF around the neutral axis of the beam.Additionally, also the difference I nF − I PZT is included With these definitions, the coupling terms can be expressed as This result can be used to discuss different geometries and types of bimorphs.Obviously a beam that consists only of piezoelectric material does not have any coupling at all, because the distance e is exactly one half of the thickness of the piezoelectric layer, e = − ℓ 3 2 .T h i s means the term I nF − I PZT vanishes.Contrary to this, a beam which is made of two identical piezoelectric layer is represented by e = 0 because of the symmetry, and a coupling exists.However, yet more efficient is the design of bimorphs or trimorphs with a substrate layer, which moves the neutral axis away from the surface of the piezoelectric layer.This results in a positive value e > 0. The best type is a symmetric trimorph with identical piezoelectric layers on both sides of the substrate layer.Here the distance equals half of the substrate layer.More details about the optimization of bimorphs can be found in [10].
In general, the piezomechanical system can be described by the following differential equations, with the modal mass m mech , the external force F(t) and the voltage u p (t) at the electrodes of the piezoelement.

Linear energy harvester
Based on these results, the simplest and linear energy harvester can be modeled.In this case only a resistor is connected as an electrical load at the electrodes of the piezoelement.Therefore the voltage u p is dependent on the time derivative of charge Qp and the differential equations of motions for this case including damping d mech read We seek for the amplitudes of the stationary oscillations, with the corresponding system matrices according to Equation 19.The amplitudes of the time signals are marked by a hat.With the stationary charge amplitude Qp the instantaneous power p(t) can be calculated, where i p is the current.The energy that is dissipated in the resistor will be treated as the stationary harvested energy E h,stat .It is then the integral of the power p, This result, normalized to force amplitude, is shown in Figure 3 versus the load resistance R and the excitation frequency Ω, normalized to the mechanical eigenfrequenzcy ω 0 .F o rt h i s study, the following system parameters are used: Obviously the harvested energy is highly frequency dependent.Only in a narrow frequency range around the eigenfrequency the efficiency is high.Also the resistor must be tuned for maximum harvested energy.The system is less sensitive towards changes of the resistance, but one can show that the optimal resistance which is optimal for low coupling and/or high damping is obtained as

Array configuration
Utilizing an array configuration for the extension of the generator bandwidth is a commonly used approach.The idea is simple and powerfull: multiple generator elements are tuned to slightly (a few %) different eigenfrequencies.A major factor is the connection to the electrical system; if each individual element uses its own bridge rectifier, the elements are electrically uncoupled and their output power simply can be summed up.However, individual bridge rectifiers are a comparable big effort, a huge number of elements and cables is needed, further the voltage drop over each diode is also summed up, so that the losses increase.As soon the individual elements are made of one part the electrodes are connected naturally, which results automatically in parallel connected elements.In such a configuration, Figure 4, the elements are electrically coupled.This section investigates the effect of the electrical coupling on the performance of piezoelectric energy harvesting generators.Two configurations are investigated, the electrical parallel connection as well as the electrical serial connection.Both cases are utilizing two elements as most basic version of a piezoelectric array.The model is based on the linear generator model in Equation 19, therefore a fore excitation is assumed.We further assume that the applied fore is equal on all elements, representing a common support.The resonance frequency tuning is made by an adoption of the modal mass of the elements (change of tip mass), all other parameters are assumed to be constant.The configurations including boundary conditions is given in Figure 4.Each element can be represented by the linear Equation 19.In the parallel configuration the two voltages are equal and u P 1 = u P 2 = R L (i P 1 + i P 2 ) apply due to Kirchhoff's rules.Using both, the parallel configuration is described by The coupling between the elements is obviously seen in the damping matrix.In the series configuration, the current at both generators is equal.Therefore 311 Modeling Aspects of Nonlinear Energy Harvesting for Increased Bandwidth Here the coupling is evidently in the stiffness matrix.Where all parameters correspond to the ones from the modeling section.The mass for frequency adoption is Δm = 0.5g.F o rt h e steady state the system can be solved and the transfer functions can be determined analog to the single system.The total dissipated energy at the resistor is again given by Evaluating this equation for voltage and current over the load resistor gives the gained energy for the application, the result is depicted in Figure 5 for serial configuration and in Figure 6 for parallel configuration.For the in series connected generators the bandwidth is widened at high impedance loads and it is not significantly changed for low impedances.For the parallel configuration, a bandwidth expanded at low impedance loads and also not changed for high impedances.To explain why there is no widening of the bandwidth in serial connection at low impedances the Equation 27 is evaluated for both elements, using the individual voltage and the common current.Figure 5 show that the second element works as energy sink for low impedances, with the consequence, that the energy generated by element one is used to actuate the other one.Figure 7 shows this effect.Even with the overall maximum displacement of element two at low impedances the gained useable energy is low.The same effect but for high impedances shown in Fig. 6 for the parallel configuration.In this case element one is the energy sink.
Concluding, the utilization of arrays with serials or parallel electrical coupling has only a major positiv effect on the bandwidth for a matched impedance, for unmatched impedances the coupling can be a drawback.The bandwidth can be enlarged with any further element, the mean energy output over bandwidth in general is higher if the mistuning of the resonance frequencies is smaller.For power and bandwidth comparison of generator arrays it is reasonable to keep the volume of active material constant.

Switching networks (SSHI)
An important technique to enhance the energy extraction is to use nonlinear switching networks.In detail, the 'Synchronized Switch Harvesting on Inductor' (SSHI) technique is studied.Such networks are an active field of research [3,4,7].The corresponding network is shown in Figure 8.This nonlinear electric circuit consists of a switching LR-branch, a rectifier and load capacitor C r .The load is again described as a resistor R L .Assuming a sinusoidal mechanical deformation of the piezoceramics, the switch is briefly closed on minima and maxima of the deformation.During these times, an oscillating electric circuit is formed, as the capacitive piezoelectric transducer and the inductance are connected.During this electrical semi-period the switch is kept close and the voltage is inverted.As this electrical period time is generally much shorter than the mechanical one, this occurs nearly instantaneously.After inversion, the switch is opened again until the next deformation extremum.Consequently, the resulting voltage signal at the piezoelectrodes is nearly rectangular-shaped.Previous publications have proven the enhanced performance of SSHI circuits especially for systems with low piezoelectrical coupling.
The modeling and optimization of such networks is not straight forward, as the overall system is nonlinear.In the following we will present a modeling technique that is based on the harmonic balance method.

Period response for harmonic excitation
Firstly, the periodic response if the SSHI is studied.In order to simplify the results, the following approximations are defined, and the electrical losses remain constant.In practical realizations, all approximations are appropriate.A small value results in a fast inversion, and a large storage capacitor means that the voltage at the load is nearly constant.Both situations are typically wanted.Further on, the nonlinear system can be treated as a piecewise linear system.
In order to obtain the stationary voltage signal it is necessary to study one semi-period of the system and consider the stationarity condition, which means the signal repeats after each period.W edefinethetimeaxisinsuchawaythatfort = 0 the voltage was just inverted and the switch is opened.With the approximation C r → ∞ the voltage u L at the load capacitor is constant at u 0 .This value has to be calculated yet.Because of the loss resistances -described by the electrical damping ratio ζ -the voltage changes from ±u 0 to ∓u 0 e −πζ .That means the absolute value of the voltage u p at the piezoelectrodes after inversion is smaller than the voltage at the load u 0 .Therefore the recifier blocks, and the piezovoltage changes linearly with the piezodeformation q, with an integration constant C 0 .According to the definitions, the piezodeformation must have an extremum at t = 0, so that the time signal reads and for the voltage signal it follows The voltage amplitude u 0 is still unknown, but at least the time t 1 can be calculated, at which the voltage at the piezoelectrodes equals the voltage at the load, u p (t = t 1 )=u 0 : At this time t 1 the rectifier changes from 'blocking' to 'conducting'.Practically this means that the piezovoltage remains constant at u 0 from t 1 until the end of the semi-period T/2.
The piezovoltage u 0 can be calculated based on the energy balance.Therefore, the transferred energy E t , the harvested energy E h and the energy dissipated in the switching branch E s must be obtained.They read, respectively, The transferred energy corresponds to the total energy that is shifted from the mechanical system into the piezoelectric system, while the other terms are the energies that are dissipated within the load resistor and the resistor of the switching branch.In stationary situation, the equality holds.With this equation, finally the stationary voltage amplitude can be recalculated as Figure 9 shows the time signal of the voltage u p .Inserting the stationary voltage amplitude into the energies in Equation 33 gives us the stationary energies, All these energy terms have quadratic dependency with the force factor α,w h i c hs h o u l db e maximized for a high energy conversion.Also the electrical damping ratio ζ in the switching branch should be low.They also grow quadratically with the vibration amplitude q of the oscillator.
However, for the force excited vibrations, the vibration amplitude is influenced by the harvesting device.The transferred energy E t,stat yields a damping effect upon the oscillator, which reduces the vibration amplitudes.In order to determine the vibration amplitudes, the harmonic balance method is applied in the following.In this technique the shunted piezoceramics is replaced by a spring -damper combination.Therefore the period -but not harmonic -force response F p (t)=−αu p (t) of the shunted piezoceramics is expressed by its Fourier-series, The Fourier-coefficients a i , b i are obtained by the periodic timesignal u p (t), The idea of the proposed linearization techniques is to approximate the periodic voltage signal by its main harmonics, This harmonic force signal is also produced by a spring -damper combination with the following parameters, In general, these replacement parameters c * , d * are frequency dependent.With these results, the stationary vibration amplitudes of the oscillator with shunted piezoceramics can be recalculated, Finally this stationary vibration amplitude q can be inserted into Equation 36 for the harvested energy E h,stat .The resulting energy is shown in Figure 10 versus the excitation frequency Ω and the load resistance R.This figure can be compared with the linear resistance case in Figure 3.It again shows that most energy is harvested at the resonance frequency ω 0 of the oscillator, because the vibration amplitudes are highest in this case.But also the load resistor must be tuned correctly in order to achieve the maximum energy.Compared to the standard case the maximum amount of harvested energy is similar, while the frequency bandwidth with SSHI tends to be larger.However, the voltage at the load resistance with SSHI circuit is nearly constant which is wanted in most practical cases, while it is a harmonics oscillations with the linear resistance.Additionally the damping effect upon the mechanical structure is higher, because of the additional energy that is dissipated within the resistance of the switching branch.

Piezomagnetoelastic energy harvesting
One major drawback for energy harvesting systems is that conventional generators produce the maximum energy when the system is excited at its resonance frequency.If the excitation frequency shifts the output power is drastically reduced.To overcome this disadvantage multiple researchers work on different broadband techniques to widen the operational frequency range.The focus in this section is on piezomagnetoelastic energy harvesting strategies.The equations of motion (EOM) are derived in the previous section.The generator is based on a model with lumped parameters.The broadband response is achieved by using nonlinear magnetic forces.Piezomagnetoelastic generators are studied in a bunch of multiple research activities [1,2,13,14].In many approaches the system is modeled as Duffing oscillator.Usually the system parameters in the model are adjusted manually to match the amplitude or power response of the experiment.The aim in this work is to investigate an analytically approach to derive the duffing parameters out of the system parameters.This section is organized as followed.
The mechanical EOM is derived for the piezomagnetoelastic energy harvesting system.In the following the duffing parameters are derived with respect to the system and input parameters and the system dynamic is discussed.The last part shows the analytic solution for large orbit oscillations.

Modeling of the piezomagnetoelastic energy harvesting system
Figure 11 gives a schematic view of the piezomagnetoelastic system.The energy harvester consist of a cantilever with two piezoelectric patches mounted on each side of an inactive substructure.The system is excited by a harmonic force where F is the amplitude of the excitation and ω is the excitation frequency.A magnetic tip mass is attached to the free end of the beam.Another permanent magnet is stationary mounted near the free end.The magnets are oppositely poled so they exhibit a repulsive force.
The nonlinear magnetic force leads to two stable equilibrium positions.Figure 11 shows both symmetric stable equilibrium positions.The tip displacement is given with q and the magnet spacing is s.The coupled mechanical and electrical differential equations are derived in the previous section and Equation 43 and 44 are similar to 19 where m = m mech is the modal mass.The modal damping is d and c = c mech + α 2 C p κ 2 is the total mechanical stiffness.Additionally to the linear differential equations the nonlinear magnetic force leads to where μ 0 is the permeability of free space, µ A and µ B are the magnetic dipole moment vectors.r AB is the vector from the source of magnet B to magnet A. • 2 is the EUKLIDEAN norm and ∇ is the vector gradient defined with The magnetic dipole moment vectors are written as where M A and M B represents the vector sum of all microscopic magnetic moments within a ferromagnetic material and V A and V B the volumes of the magnets.Details of the force between magnetic dipoles can be found in [16].φ is the rotation angle at the magnet A.
The vector from the source of magnet B to magnet A is where l A and l B are the length of the magnets and l P is the length of the beam in x-direction.
where U mech is the mechanical potential energy and c is the equivalent stiffness of the beam.
U is normalized to the potential energy U 0 for a magnet distance of s = 0.79s 0 .T h e t i p displacement q is normalized to the tip displacement q 0 for s = 0.The derivation of the mechanical potential energy expression is shown in [15] and is proportional to q 2 for the first bending mode.U mag is given in Equation 45.The Figure shows the potential energy for different magnet distances with respect to s 0 which is the critical magnet distance where the potential energy change from two to one stable equilibrium position.The nonlinear magnetic 0.66s 0 0.79s 0 1.00s 0 1.31s 0 q/q 0 q/q 0 U/U force leads to an energy hump at zero tip displacement.A sharp peak results for a small spacing.
The system exhibits two stable equilibrium positions which are the local minimum positions in the potential energy and one unstable equilibrium for zero displacement (s < s 0 ).The hump disappears for large s and there is only one stable equilibrium for zero displacement (s > s 0 ).
The normalized restoring force is additionally presented in Figure 12 b).F res is normalized to the restoring force for a magnet spacing of s = 0.79s 0 .Equation 50 is a nonlinear function of the system parameters in particular a function of s.
The benefit of the magnet force is the nonlinearity in the system response.Due to the restoring force the system exhibits overhanging resonance curves which strongly depends on the parameters.For s < s 0 and low excitation energy that the system only oscillates around one equilibrium the EOM has a hardening stiffness so the resonance curve overhang to the left.A softening response is given for s > s 0 with a distorted peak to the right.Specially for a hardening stiffness the system exhibit two equilibrium positions and the system bounces between both positions if the energy of the excitation is high enough.These large orbit deflections generate the most energy and are most important for designing the system setup.

Approximation with the Duffing oscillator
The potential energy shown in Figure 12 a) is approximated as fourth order polynomial function in a bunch of different research activities.The nonlinear restoring force shown in Figure 12 b) is than approximated as the Duffing equation with the Duffing parameters α and β.The approach as Duffing equation is not only limited to piezomagnetoelastic energy harvesting techniques.[5] used this equation to model an electrostatic energy harvesting system and [8] approximate an electromagnetic energy generator.In the following α is the Duffing parameter before the cubic term instead of the coupling factor in Equation 43 to be conform to the most publications concerning the duffing equation.The uncoupled equation of motion (EOM) for the piezomagnetoelastic system in terms of the duffing oscillator is written as The Duffing parameters are complicated functions of the beam, the piezo, the magnet parameters and in particular the magnet distance.It can be recognized in Figure 12 b) that the cubic part αq 3 is always positive for repulsive magnets because the function is always monotonically increasing for large q values.Only the linear part βq can be either positive or negative with respect to the magnet distance.For s < s 0 the parameter β is negative and the system exhibits two stable and one unstable equilibrium position.For s ≥ s 0 the system has only one stable equilibrium position and β is positive.s 0 is the critical magnet spacing.
The nonlinear restoring force F res shown in Figure 12 b) can be approximated as a third order Taylor serious given with p(q)=k 1 q 3 + k 2 q 2 + k 3 q + k 4 (54) where k 1−4 are constants derived from the system parameters.By comparing Equation 54 and Equation 52 note that k 1 = α and k 3 = β.To calculate the constant parameters in Equation 54 it takes four constrains.
1.The restoring force is point symmetric: 2. The Taylor series must exhibit the same equilibrium positions: p(q eq )=0. (57) 3. The oscillation must be suitable for small magnet distances s around one equilibrium position dp(q eq ) dq = a, ( where q eq is the equilibrium position and a is the slope at q eq .a and q eq can be calculated from the original system.The solution for k 1−4 is ) Equation 53 becomes Equation 60 is usable if the potential energy exhibits two equilibrium positions.If the magnet distance is larger than s 0 the Duffing parameters α and β are simply the liner and the cubic part of the third order Taylor series of Equation 50 and α and β are positive.
This behavior can be recognized in Figure 13 a) and b).The graphs are calculated by using the potential energy of two point dipols in Equation 49. Figure a) shows the equilibrium position over the magnet distance.The equilibrium position q eq is normalized to the known q 0 .The magnet distance is normalized to the critical magnet distance s 0 .Note that the distance between the two stable q eq becomes smaller for very small s because of the rotation angle φ of magnet A. The Figure 13 b) gives the modification of the Duffing parameter β with respect to the normalized known magnet distance.β is normalized to β 0 the linear Duffing parameter Duffing q/q 0 ω/ω 0 q eq q eq q eq q eq −q eq −q eq s = 0.38s 0 for s = 0.For s < s 0 the system exhibit a negative linear restoring force due to the negative parameter β. β is positive for s > s 0 .
Figure 14 shows the approximation of Equation 50 with Equation 52 and the Duffing parameter known from Equation 59.The pictures a) show the restoring force derived from the force between two point dipoles and the Duffing equation for different magnet distances.The pictures b) gives the corresponding time variant tip displacement.It can be recognized that the Duffing equation approximate the restoring force very well for the different ranges of magnet spacing.The Duffing equation shows the hardening response for small magnet distances (s = 0.38s 0 ) and it approximates the influence of the different attractors (s = 0.91s 0 ).For s > s 0 the approximation is also very good.

Solution for large orbit oscillations
In Section 5.1 and Section 5.2 the piezomagnetoelastic energy harvesting system is presented and an approximation with the Duffing oscillator is given.In the following Equation 60 is solved for large orbit oscillations so that system bounces between both symmetric equilibrium positions or around the only one for s > s 0 .The EOM can be written as q + 2Dω 0 q + ǫq 3 + sgn(β)ω 2 0 q = f 0 cos(ωt) (61) where sgn(β) is either positive or negative for β > 0orβ < 0with The Equation 61 can be solved by applying the harmonic balance method.The harmonic balance method is well known and details can be found in [12].The amplitude response is assumed as harmonic with the frequency of the excitation q = qcos(ωt − ϕ) (63 where q is the amplitude and ϕ is the phase of the tip displacement.Insert Equation 63 in Equation 61 and only consider terms with the excitation frequency leads to sgn(β)ω 2 0 − ω 2 q + 3 4 ǫ q3 2 + 4D 2 ω 2 0 ω 2 q2 cos (ωt − ... ...− ϕ + arctan 2Dω 0 ω q (sgn(β)ω 2 0 − ω 2 ) q + 3 4 ǫ q3 The Equation 61 is valid if the amplitude and the phase in Equation 63 solves the equations.The solution gives the frequency response with respect to the amplitude q Figure 15 shows the analytical amplitude response given in Equation 65 and the numerical solution of the Duffing oscillator.The graph shows the solution for three different magnet distances.The frequency was slowly increased so the system remains a steady state response.One can recognize that the harmonic balance is well suited if the Duffing equation has a positive linear restoring force (β > 0).If the Duffing oscillator exhibits a negative restoring force and the system excitation delivers enough energy that the energy harvester bounces between both stable equilibrium positions than the harmonic balance predicts the influence of Numerical Analytical s = 1.18s 0 , β > 0 s = 1.05s 0 , β > 0 s = 0.98s 0 , β < 0 ω/ω 0 q/q 0 q/q 0 q/q 0 0 0.5 the attractors till the first jump.The system behavior after the first jump can not be predicted with this harmonic approximation.

Conclusions
This chapter presents different techniques to enlarge the frequency bandwidth of piezoelectric energy harvester.A precise modeling of piezomechanical structures is given, and the linear harvesting system is given as the reference.In detail a nonlinear switching SSHI circuit, nonlinear magnet forces and an array configuration of several bimorphs are discussed.For all cases, appropriate modeling techniques are presented that allow an efficient yet precise analysis.The nonlinear techniques alter the system dynamics especially for off-resonance vibrations.The magnet forces generate a bistable system, in which the bimorph oscillates between both equilibria.The SSHI circuit increase the energy conversion rate by producing a rectangular shaped voltage signal, while the array configuration tunes the individual bimorphs to slightly different frequencies, so that the energy conversion is distributed to a broader frequency range.

Figure 3 .
Figure 3. Harvested energy E h,stat versus normalized excitation frequency Ω and load resistance R.

Figure 9 .
Figure 9.Time signals of piezodeformation and voltage at the electrodes.

Figure 10 .
Figure 10.Harvested energy E h,stat versus normalized excitation frequency Ω and load resistance R.

Figure 11 .
Figure 11.Schematic view of the piezomagnetoelastic energy harvesting system

Figure 12 a
Figure 12 a) presents the potential energy

Figure 12 .
Figure 12.Potential energy and restoring force

Figure 13 .
Figure 13.Equilibrium positions and β with respect to the magnet distance s.

Figure 14 .
Figure 14.Duffing approximation of the restoring force and displacement

Figure 15 .
Figure 15.Numerical and analytical results for the large orbit duffing equation