ON THE COMPARISON, SCALING AND BENCHMARKING OF ELECTROMAGNETIC VIBRATION ENERGY HARVESTERS C. Cepnik* and U. Wallrabe University of Freiburg – IMTEK, Department of Microsystems Engineering, Laboratory for Microactuators, Freiburg, Germany *Presenting Author: Clemens.Cepnik@imtek.de Abstract: This paper introduces and compares different benchmark parameters for the performance of electromagnetic energy harvesters, including the bottom-up (power according to the harvester model) as well as the top-down view (maximum power that can be generated within a volume) and considering the dependency and scaling of geometrical and physical parameters, such as the mechanical damping. The comparison is not only to give assistance for a fair benchmarking of prototypes but moreover to better understand how high performance harvesters should be designed. We summarize, which important data should be provided with a published harvester to allow benchmarking, and conclude to a set of benchmark parameters that are of main practical importance. Keywords: electromagnetic vibration energy harvesting, performance, benchmarking, scaling, damping INTRODUCTION The electromagnetic energy harvesters of different shape and dimension that have been published for more than a decade raise the question about what a good design is like. The difficulty to answer the question is, firstly, the complex impact of boundary conditions, such as the volume and the excitation, on the output power, secondly, the different boundary conditions and fields of application harvesters have been designed for, and, finally, the few data published for most prototypes. Different authors compared electromagnetic harvesters and introduced benchmark parameters (BPs). While Mitcheson et al. [1] applied the top-down approach, i.e., referred to the maximum power that can be generated within a certain volume, Arnold et al. [2] came bottom-up and investigated a simplified linear model as well as published harvesters to assess basic scaling of the output power. Based on the same model O’Donnel et al. [3] discussed further fundamentals of harvester scaling. We add a broader overview by a) accounting for single physical parameters and b) investigating scaling of physical as well as geometrical parameters and their impact on the output power. In addition to discussions of the electrical damping in [4] we will review the impact of the mechanical damping on the output power, i.e., damping by drag, Stokes and Couette damping, squeeze film damping, anchor loss and thermoelastic damping. We finally derive advanced BPs, provide experimental examples and discuss the parameters with respect to practical importance. APPROACHES FOR NORMALIZATION The maximum average power that can be generated within a certain volume and was first introduced by Mitcheson et al. [1]. If one assumes that: The harvester volume is completely used for the linear oscillator and the oscillator displacement, the mechanical damping as well as the volume for coils, casing, airgaps etc. are neglected and the oscillator is only damped electrically, for the optimum oscillator vs. harvester height of 0.5 the average power becomes Pavg = 1 ωρm V ha 8π (1) with V the harvester volume, a the acceleration, ω the angular frequency and ρm the mass density. On the other hand, to describe the behavior of a resonant harvester including the mechanical damping dm the output power often is derived from the normalized model [3] to Pavg = m2 a 2 m2 a 2 = 4 (dm + de ) 8 dm (2) with m the oscillator mass and de the electrical damping. Here, due to the normalization, the optimum de equals dm, because the efficiency of the electrical circuit is not considered. Nevertheless, equation (2) mostly is a good approximation. In contrast, we derived in [5] the output power without normalization to better account for the physical and geometrical parameters and found Pavg = V c kc B 2 m2 a 2 8dm (Vc kc B 2 + ρc dm ) (3) where Vc is the coil volume, kc the copper filling factor, ρc the resistivity and B the flux density perpendicular to the wire and the direction of motion. The difference compared to equation (2) becomes obvious with the relation of de and dm [5]. Rl + Rc dm = de Rl − Rc (4) At high flux gradients and low dm the load resistance Rl potentially is a magnitude larger than the coil resistance Rc, by which Rl and Rc become nearly equal [5]. SCALING Based on eqs. (1)-(4) the scaling of the output power with respect to the length scaling factor s can be estimated. One needs to consider a) the environmental boundary conditions: excitation frequency and acceleration, b) the harvester parameters: aspect ratio and total volume and c) the physical limitations: mechanical and electrical damping. In contrast to a) and b), the impact of the damping requires a more detailed investigation. Electrical damping For the electrical damping de = l2 B 2 Rl + Rc (6) we find with the optimum load according to [5] de = dm V c kc B 2 Vc kc B 2 + 2ρc dm (7) B does physically not scale but can decrease at miniaturization due to fabrication limitations and large relative tolerances [4]. At high VckcB2 and low dm (8) de ∝ dm while for low VckcB2 and large dm (9) de ∝ V c ∝ s 3 In case Rl = Rc, de in eq. (6) scales equally to (9). However, the assumption Rl = Rc is only meaningful when either the vibration amplitude needs to be reduced (by increasing Lorentz force) or a power management circuit with a similar input resistance is used. However, in these cases the design should be changed to allow a larger amplitude and respectively to realiue a lower Rc. Thus, eqs. (8) and (9) apply. Mechanical damping Mechanical damping has been broadly investigated for sensors and equally appears for energy harvesters. Its scaling is more complex compared to the electrical damping, as different models apply depending on the geometry. A detailed comparison of different designs with different aspect ratios requires extensive measurements or simulations and is beyond the scope of this paper. Our discussion is based on a review, how dm scales according to the most relevant models. a) Viscid drag damping. Drag damps a moving body surrounded by a fluid in the laminar flow region. The damping coefficient can be calculated to dm = 6πµr ∝ µs (10) r is the representative dimension and µ is the viscosity. b) Viscid lateral motion damping. Assume a small gap g between a plate and a surface. The plate with the surface A (dimensions much larger than g) moves parallel to the surface. At a fully developed linear time-independent flow one can model the damping of the plate as Couette damping [6] and gets dm = µA/g ∝ µs (11) In case the time dependency of the flow is considered the damping is called Stokes damping [6]. The viscid damping by the fluid between plate and surface then is ! 1 √ (12) ωρµ ∝ β ωµ s2 2 -1 with β = f(ωg), β ∝ 1 for large ωg and β ∝ s for small dm = βA ωg [6]. Thus, at very thin gaps and/or low frequencies Stokes equals Couette damping. While the boundary conditions in the Couette model neglect the force onto the plate top surface, in the Stokes model this damping component is considered by an infinite gap [6]. ! dm = A 1 √ ωρµ ∝ ωµ s2 2 (13) For both, Couette and Stokes damping, edge and size effects that appear at finite lateral dimensions only add a factor but do not change scaling [7]. c) Viscid perpendicular motion damping. Squeeze film damping occurs when a plate oscillates perpendicularly to a surface. For a small vibration amplitude and a small characteristic squeeze film number, which means that gas can flow out of the gap and is not trapped within [8], dm is given by [9]. dm = µLw3 /g 3 ∝ µs (14) In case of large amplitudes Starr [10] found an additional factor of (1-ε)-3/2 with ε the vibration amplitude in relation to g. Again, edge and size effects do not change scaling [11]. Viscid damping is more complex at high velocities as higher order damping terms have to be considered. According to [14] for turbulent drag follows dm ∝ ρs2 ẋ (15) With the Reynolds number proportional to s, at large volumes and excitations dm approaches s2 x˙ . All viscid damping models depend on the ambient pressure and can be reduced at rarefaction. According to [11,12] different pressure regions are distinguished ! depending on the Knudsen number Kn = λ 0 P0 gP (16) with λ0 the mean free path at a known pressure P0 and P the operating pressure. The first region, the slip flow regime is defined at 0.01 < Kn < 0.1. Regions of lower pressure are the transition (0.1 < Kn < 10) and the free molecular regime (Kn > 10). Since a Kn of 0.1 in air does already correspond to a representative physical length (or airgap) of only 0.68 µm P0/P, which by now covers all harvester prototypes, we limit the review to the slip flow regime. Here, the impact of P is considered by the effective viscosity µeff, which was found empirically [11,13] and slightly differs depending on the publication. In each case dm is reduced at lower pressures. d) Other types of damping. Damping through sound waves generated by the oscillation is called acoustic damping [11] and is usually neglected. Anchor loss or loss through imbalances is due to tensions coupled into the clamping and can have a significant impact at low pressures [11,15,16]. However, it is neglected for pressures down to the slip flow regime. Thermoelastic damping [15,17] is caused by a temperature difference that appears when tensions change in a material, e.g., in the harvester springs. Also, similarly to anchor loss damping, it can be neglected. Imperfections in the material of the springs can additionally cause heat dissipation when chemical or physical bonds are made or broken. This internal friction is neglected at even lower pressures [15]. Table 1. Possible benchmark parameters with possible field of application and practical limitations. (a) (b) (c) (d) (e) (f) Benchmark P/V P/(1/4 fρVha) P/(V1/3a) P/(V5/3 a2) P/(V7/3a2) P/(V2a2) Possible field of application volume comparison to batteries / other mechanisms max. performance of different harvesting principles el.-mag. harvesters in turbulent flow el.-mag. harvesters in laminar flow, high ηmag el.-mag. (micro) harvesters in laminar flow, low ηmag el.-mag. harvesters in laminar flow (simplified) 10000 Practical limitations no information about excitation max. el. damping not considered drag must be turbulent drag must be laminar drag must be laminar drag must be laminar, simplified [20] [19] [22] [21] [5] [20] [19] [22] [21] [5] [20] [19] [22] [21] [5] Resoance frequency 1000 (*) 100 (*) (*) commercial harvesters, no data about design published 10 [18] 0.001 [18] 0.1 10 [20] [19] [22] 0.1 1000 [5] 1000 [18] 10 1000 [20] [19] [22] [21] [5] 0.1 10 [20] [19] [22] 1000 Volume [5] (*) 100 [21] 10 [21] [18] 1 [18] [18] Figure 2. Benchmark plots for published electromagnetic energy harvesters. Value proportional to area of circle. Thus, for benchmarking energy harvesters it is appropriate to only consider viscid and electrical damping as shown in figure 1. Due to the high nonlinearity viscid damping can thereby only be modeled for singular regions of volume, excitation and pressure. The complexity and the lack of a comprehensive theoretical model require to define a set of BPs with respect to different boundary conditions and matters of interest. For a lack of space and harvester prototypes at reduced pressure we limit the discussion to standard pressure conditions. Damping [a.u.] 1E+04 1E+02 turbulent drag Stokes damping 1E+00 el.-mag. damping laminar drag, Couette & squeeze film damping 1E-02 1E-04 0.01 0.1 1 10 100 Scaling factor s [a.u.] Figure 1. Scaling of damping with respect to volume. BENCHMARKING Different benchmark parameters (BPs) as shown in table 1 can be derived from practical aspects and theory. An electronic circuit could require a certain maximum voltage or power and certain average values. However, we do not consider these values, because they can easily be changed by volume scaling or changing the boundary conditions. The power density (BP (a)) is useful for a comparison with batteries and other harvesting principles but does not consider the practical dependency on the influencing variables. According to equation (1), with ρm equal for all harvesters, the maximum power (b) that can be generated in a certain volume on the one hand enables to compare different harvesting principles. On the other hand it does not consider mechanical damping and the maximum possible electrical damping, which both are the main reasons why the output power in practice is much lower than one would expect from equation (1). From eq. (2), which accounts for the mechanical damping, we find the laminar scaling (d) in eq. (17) assuming dm ∝ s (compare [2]). Pavg ∝ s5 a2 ∝ V 5/3 a2 (17) 2 ∝ In contrast, for dm s x˙ the turbulent scaling (c) results when a nearly sinusoidal motion is assumed. Pavg ∝ sa ∝ V 1/3 a (18) The difference!between eqs. (2) and (3) is the factor ηmag = V c kc B 2 Vc kc B 2 + ρc dm (19) which we call effectiveness of the magnetic circuit. For benchmarking two scenarios are of general interest: 1.) For microharvesters low ηmag and laminar damping, i.e., dm ∝ s, are realistic. 2.) For larger harvesters the design freedom allows ηmag à 1, laminar as well as turbulent damping should be considered. For the first case BP (e) follows (compare [2]). Pavg ∝ s7 a2 ∝ V 7/3 a2 (20) Scaling in the second case equals eq. (17) and (18). Finally, the difference between BP (d) and (e) is small. In case it is difficult to decide which of both is more appropriate, we suggest to use the average BP (f) Pavg ∝ s6 a2 ∝ V 2 a2 (21) which interestingly results from eq.(2) with dm = const. Figure 2 contains the BPs for published el.-mag. harvesters with respect to their volume and resonant frequency. If measurements at different excitations are available for one prototype we use the best result in each diagram. The plots show that only a few harvesters are high performing according to all BPs. Available and estimated key parameters of best performing prototypes are collected in table 2. Considering the design of these harvesters shows that different approaches are target-aiming. A low performance mostly is due to a small effectiveness of the magnetic circuit per total volume. In case a prototype is only high performing according to one BP reasons could be: - the harvester has a small effectiveness of the magnetic circuit but profits from a certain BP - measurements are only available for a certain excitation, at lower / larger excitations the performance can increase according to (d) / (c) - a low aspect ratio increases (a) but decreases the other parameters and vice versa - the concept is different to the standard oscillator (e.g., a pendulum or frequency up-conversion) For our cylindrical prototype figure 3 shows the BPs and mechanical damping vs. the acceleration. This plot indicates that BP (d)-(f) are appropriate for laminar drag and (c) for turbulent drag. Table 2. Key parameters of high performing harvesters in SI units computed or cited / estimated / simulated. Vc/V [18] [19] [20] [21] [22] [5] 0.013 0.020 0.020 0.050 0.077 0.053 kc Bavg dm ρm xmax/h m/V 0.15 0.50 1.00 0.31 0.02 0.29 0.36 0.41 0.33 0.64 0.0130 0.0046 0.2-0.4 ~7600 ~7800 ~16000 ~7000 ~7800 ~7800 0.13 < 0.04 0.40 < 0.02 1100 2100 3500 1000 2900 3650 1E05 1.0 1E04 [a.u.] Benchmarks [a.u.] *[18]: frequency up conversion, [21]: circulating pendulum 1E03 (d), (e), (f) 1E02 1E01 (a) 0.2 (b), (c) 4 2 1 Acceleration 0.2 laminar damping 0.1 0.2 2 1 4 Acceleration Figure 3. BPs for the measured power and damping (open circuit) of our harvester [5] vs. acceleration. For small accelerations laminar damping dominates, for accelerations > 1ms-2 damping increases. CONCLUSION Since it is impossible to find a benchmark parameter (BP) that accounts for all boundary conditions at different magnitudes and since different energy sources scale differently, we can conclude that depending on the focus of research or the potential application different BPs need to be used: 1. (a), (b) to compare different harvesting principles 2. (c) to benchmark el.-mag. harvesters with turbulent flow, i.e., large volume or strong excitation 3. (d) to benchmark el.-mag. harvesters with laminar flow, i.e., small excitation and very small volume 4. (e) to benchmark el.-mag. harvesters with laminar flow, i.e., small excitation but larger volumes 5. (f) to benchmark both, (d) and (e). To enable a better harvester characterization and design evaluation measurements with different boundary conditions should be provided. This especially includes measurements of the 1. output power, 2. output voltage and 3. vibration amplitudes (open circuit and with Rl) at different acceleration amplitudes. To compare the effectiveness of the magnetic circuit geometrical and material parameters of the design and the prototype should be provided. We encourage the reader to provide as many data as possible to contribute to the challenge of finding most beneficial harvester designs. Finally, from the application point of view, additional parameters are of broader interest: robustness against temperature, tolerated peak acceleration and lifetime. ACKKNOWLEDGEMENTS The first author thanks P. Mitcheson and E.M. Yeatman for the fruitful discussions about [1]. REFERENCES [1] [2] [3] [4] Mitcheson P.et al.,J.Micromech.Microeng.(2007) 17,211-216 Arnold D.P., IEEE Trans. Magn. (2007) 43, 3940-3951 O’Donnel T. et al., Microsyst.Technol. (2007) 13, 1637-1645 Cepnik C. et al., Proc. PowerMEMS’10 (Leuven, Belgium, 2010) 69-72 [5] Cepnik C. et al., Sens. Actuators, A (2011) 167, 416-421 [6] Cho Y.H. et al., Proc. MEMS’93 (Fort Lauderdale, FL, USA, 1993) 93-98 [7] Zhang X. et al., Proc.MEMS’94 (Oiko, Japan) 199-204 [8] Andrews M. et al., Sens. Actuators, A (1993) 36, 79-87 [9] Bao M. et al., Sens. Actuators, A (2007) 136, 3-27 [10] Starr J.B., Solid-State Sens. Actuator Workshop (Hilton Head Island, SC, USA, 1990) 44-47 [11] Lin R.M. et al., Mech.Syst.Sig.Process. (2006) 20, 1015-1043 [12] Newell W.E., Science (1968) 161, 1320-1326 [13] Veijola T. et. al.,J.Microelectromech.Syst.(2002) 10, 263-273 [14] Bao M.H., Elsevier (2000) ISBN: 978-0444505583 [15] Stemme G., J. Micromech. Microeng. (1991) 1, 113-125 [16] Buser R.A. et al., Sens. Actuators, A (1990) 21, 323-327 [17] Zener C., Phys. Rev. (1937) 52, 230-235 (LiW06) [18] Zorlu O. et al., IEEE Sens. J. (2011) 11, 481-488 [19] El-Hami M. et al., Sens. Actuators, A (2001) 92, 335-342 [20] Beeby S.P. et al., Microsyst.Technol. (2007) 13, 1647-1653 [21] Spreemann D.et al.,J.Micromech.Microeng.(2006) 16,169-173 [22] Cepnik C. et al., Proc. Actuators’11 (Beijing, China, 2011) 661-664