Floor Tile Energy Harvester for Self-Powered Wireless Occupancy Sensing

Analytical

We begin our analysis by first defining the schematic representation of the cymbal transducer, shown in Figure 2(a). The vertices of the angle bends in the end caps are labeled A-F so we may identify the individual segments of the cymbal by the lines connecting the two points. Distributed load, P, is applied to section BC, and section EF is considered to be resting on a rigid surface. Following the assumptions of Fernandez et al. (1998) that cymbal end cap bends behave as pin connected joints and the end cap members are rigid, such that there is no energy loss due to end cap bending, we add the conjecture that the bulk piezoceramic between the end caps behaves as a stiff spring. These assumptions are represented in Figure 2(b). Therefore, each cymbal section is considered to be a simple truss, and the force distribution in the structure is solved using the method of joints. Figure 2(c) shows the sign of the force in each section, with the cymbal end caps being in compression, and the piezoceramic layer being in tension. By the method of joints,

and

Therefore,

Figure 2

Schematic representations of analytical model with cymbal end caps in black and piezoelectric material in orange, and (a) diagram of cymbal with labeled points and applied force, (b) diagram with ideal model assumptions, (c) illustration of force sign in the cymbal members, (d) member responsible for horizontal (x-axis) deflection and (e) member responsible for vertical (y-axis) deflection

It is worth noting here that the cotangent of the angle θ essentially acts as an amplification factor to the applied force, P, as for any angle θ less than 45°, the cotθ is greater than one.

With force exerted on the piezoelectric layer known, we can examine the resulting deflections of the cymbal in both the horizontal (x-axis) and vertical (y-axis) directions. The deflection in the horizontal direction is caused by the lengthening of the piezoelectric layer under tensile forces, and following our assumptions for a member with initial length, x, the deflection, dx, as shown in Figure 2(d), is equal to,

where Ep is the Young’s modulus of the piezoelectric and A is the cross-sectional area of the piezoelectric layer, which for a rectangular cymbal is the thickness of piezoelectric, tp, multiplied by its depth, b. The elongation of the piezoelectric materials in the horizontal x-direction is accompanied by a subsequent compression of the end caps in the vertical y-direction. Examining one of the slanted segments of the cymbal end cap, such as segment AB, we see that it has some initial length, R, and rectilinear components x′ and y. From our assumption of rigid cymbal end cap members, as deflection occurs, the length of segment AB remains R; however, the rectilinear components are given the addition of dx/2 and dy/2 respectively, as is presented in Figure 2(e). The deflections are halved in this case since the cymbal structure is symmetric about the center planes of the horizontal and vertical axes and we are analyzing a single end cap segment. Finally, with this information we solve for deflection in the y-direction by applying the Pythagorean Theorem to the deflected segment in Figure 2(e), where we find,

Note that for this application, where the cymbal transducer is used as an energy harvester, the quantity, dy, will always be negative.

After establishing eqs [1]–[5], it becomes evident that the performance of the cymbal transducer is largely dependent on the angle, θ, which is governed by the cymbal height, h, and inner width, di, labeled in Figure 3(a). Using the analytical model presented thus far, the plots in Figure 3(c)-(h) are generated employing the values in Table 1, with Figure 3(b) offering a qualitative illustration of cymbal cap geometry over the parametric sweep of di and h. The applied load, P, for these simulations is 80 N, which is several times smaller than an average adult’s weight, but was chosen because multiple cymbal transducers will be used inside the floor tile, which will distribute the applied load. The determination of the number of cymbals is described in the later sections of this paper.

Figure 3

Results of cymbal transducer analytical modeling with (a) notation of model parameters, (b) qualitative illustration of cymbal geometry as function of parameters di and h. Plots (c)–(h) reveal the dependency of angle θ, amplification factor, magnitude of dy, dx, stress in the piezoelectric layer and critical thickness (to buckling) of cymbal end caps, respectively. All responses are to a static 80 N load

From Figure 3(d) it can be seen that the largest amplification factors (the cotangent of Figure 3(c)) are observed where di and h become small. Naturally, as amplification factor is increased, higher force is applied to the piezoelectric layer, subsequently increasing the magnitude of displacements in both x (Figure 3(f)) and y (Figure 3(e)) directions. However, the deflection in the y-direction, which is what would be felt by the person stepping on the tile, is still significantly small, thus meeting the tile design requirements. The stress in the piezoelectric layer, plotted in Figure 3(g), is a direct result of x-direction displacement and follows the same profile as Figure 3(f). Certainly, the higher the amplifications factor, the more energy can ultimately be harvested per footstep on the tile. However, the limiting factor is the stress in the piezoelectric layer, which for the chosen material (APC 850) has a yield strength of approximately 36 MPa (APC International 2013).

In order to realize our assumption that a negligible amount of energy is lost to bending of the cymbal end caps, the end caps themselves must be as thin as possible, up to the point where they fail due to buckling, as they are under compression. The cymbal end cap members are treated as pinned-pinned plates under axial compression, as per our assumptions. The critical load for plate buckling is found using Euler’s formula,

where Pcr represents the minimum load which can be applied to a plate before causing buckling, E the Young’s modulus of the material, I the moment of inertia and Le the effective length (a multiple of the physical length, determined by the boundary conditions). For our case, a load is prescribed and we wish to find the necessary geometry to support that load. We then re-write eq. [6] as,

where the effective length is now 1−νc2a2, because we are assuming pinned-pinned joints (Rees 2009). Here νc is the Poisson’s ratio of the cymbal end cap, a is the physical length, b the width and tcmin is the minimum end cap material thickness necessary to prevent buckling due to prescribed applied load, P. Next, eq. [7] is solved for the minimum thickness,

The cymbal structure can experience buckling either on the slanted side sections or along the top member, depending on the relative values of di and h. Therefore eq. [8] makes two distinct curves as a takes the value of di to examine buckling in the top section and takes the value of R to examine buckling in the slanted side section. These two curves are plotted in Figure 3(h), where the greater of the two solutions is displayed for the varying geometry.

With this analysis, cymbal dimensions were chosen such that the stress in the piezoelectric layer would be approximately 75% of the yield strength, giving a Factor of Safety of 1.33, while keeping y-axis deflection at a minimum. The Factor of Safety is kept low to maximize the performance, and because the tile will be protected from overloading, as is described in a later section. Other considerations were to keep the height of the transducer small, such that the overall height of the floor tile could remain slim, making it easier to install in a floor, and keeping the inner width of the cymbal (di) large enough, such that the top plate of the floor tile could have a good surface to rest upon. The chosen design point is highlighted on plots of Figure 3 with a magenta marker and corresponds to di=5 mm and h=2 mm. For this geometry, the cymbal end caps would need to be at least approximately 230 μm thick. In keeping with the Factor of Safety of 1.33, the next closest available material thickness (material sold in standard thicknesses) was chosen to give tc =305 μm.

Finite element analysis

To check the validity of the analytical model and assumptions prior to fabrication, a finite element analysis of a cymbal transducer of the dimensions chosen in the previous section was conducted. The dimensions are detailed in Figure 4(a), along with the mesh view overlaid on the cymbal transducer assembly. The material properties for the steel end caps and PZT piezoelectric layer were estimated using the available material information and were identical to that used in the analytical model, listed in Table 1. The interface of the cymbal end caps and piezoelectric layer was taken to be rigidly bonded. This analysis considers 32,115 tetrahedral elements with an equivalent 80 N distributed load applied across the top surface of the transducer and was conducted using the Stress Analysis Environment of Autodesk Inventor Professional 2013.

Figure 4

Results of finite element analysis of cymbal structure subject to an equivalent 80 N distributed static load across top surface using 32,115 tetrahedral elements with (a) model dimension drawings and finite element mesh, (b) first principal stress of assembly, (c) first principal stress in piezoelectric layer, (d) vertical (y-axis) displacement and (e) horizontal (x-axis) displacement

Examining Figure 4(b), where the coloring represents the First Principal Stress, we see that the cymbal end caps are largely in uniform compression within each segment, with only a slight amount of tensile stress on the surface by the obtuse angles due to the occurrence of some finite bending in that region. Looking at the First Principal Stress in just the piezoelectric layer, shown in Figure 4(c), we find that this layer is in uniform tension, with the exception of stress concentrations in the corners caused by material necking as it meets the bonded boundary condition. The dark colored regions on either end of the piezoelectric layer are located where the end caps are bonded and are in slight compression to support the vertical component of the reaction load applied by the end caps. Vertical (y-axis) displacement of the cymbal transducer is shown in Figure 4(d) and horizontal (x-axis) displacement is shown in Figure 4(e). Note the displacement in the horizontal direction is axisymmetric and thus the displacement, dx (elongation of the piezoelectric layer), is taken to be twice the maximum value of the Figure 4(e).

A comparison of the results from the analytical and finite element models is shown in Table 2. Here we find good agreement between the two models, with the slight differences being attributed to the bending of the cymbal end caps, which is neglected in the analytical model. The bending predicted by the FEA gives additional vertical displacement, dy, which subsequently takes away from horizontal displacement, dx. This energy loss is then reflected in the slightly reduced stress observed in the piezoelectric layer. The analytical model was subsequently concluded to be sufficiently accurate and a slightly conservative representation of the performance of the cymbal transducers in this specific application.

Manufacturing

The cymbal transducers deflect only a small portion of their height (around 3%) when fully compressed. For this reason, it is necessary to develop a manufacturing technique such that the dimensions of each of several cymbal transducers in the floor tile are most nearly the same. Therefore, a press-brake manufacturing technique was adopted in order to produce a consistent cymbal end cap. The press, shown on the left, and the brake, shown on the right, in Figure 5(a), were machined in a Tormach PCNC 1100 mill in a slightly exaggerated profile to allow for spring-back in the steel end caps after pressing. Flat pieces of steel shim were inserted into the press brake and bent into the desired shape, as seen in Figure 5(b). The bending pressure was applied by a hydraulic press, shown in Figure 5(c), operating at approximately 20 kpsi. Piezoelectric elements were then packaged (discussed in a later section) as shown in Figure 5(d). The elements were arranged in five parallel 60 mm by 10 mm pieces rather than one 50 mm by 50 mm piece due to material availability. The assembly was then bonded together using Loctite 120 HP epoxy, which was allowed to cure in a kiln at 65°C for no less than 12 hours. The completed assembly is shown in Figure 5(e).

Figure 5

Manufacturing process for production of cymbal transducers whereby (a) press (left) and brake (right) are machined and used to bend (b) flat pieces of steel into shape when under pressure from (c) a hydraulic press. (d) Piezoelectric elements are then packaged and (e) the assembly is bonded together using epoxy

Testing procedure

In order to verify the performance of the cymbal transducers in the target application, it is necessary to have the ability to impose a repeatable force input onto the system. The force applied to the ground during walking varies from person to person and even between steps made by the same person. Thus, we first seek to find a mechanical means of applying force onto the cymbals so that the same force profile may be applied across multiple tests. This will allow for comparative analysis across a range of cymbal variables. To establish the requirements of the mechanical test stand, we first institute the force profile that needs to be simulated. In prior work, Martin and Marsh (1992) have provided the ground reaction force for persons walking at their preferred pace, from which the fourth-order polynomial curve fit in Figure 6(a) was derived. The curve was obtained under the assumption that the first peak (heel strike) occurs at 25% of the contact time, the trough (when both feet are on the ground) occurs at 50% of the contact time, and the second peak (toe-off) occurs at 75% of the contact time. The constraints were also set such that the first and second peaks were of the same approximate amplitude, which is true under steady-state walking, and the curve tends to zero at 0% and 100% contact time. With these constraints in place, the coefficients of the polynomial were solved as a function of T, the total contact time (i.e. period of the step), Fp, the peak force, and Fv the force in the valley of the curve as shown in Figure 6(a). The values of T, Fp, and Fv were chosen to be 0.63, 1.17, and 0.7, based on the findings of Martin and Marsh (1992). Note the ground reaction force (Vertical Force in Figure 6(a)) is given as a multiple of body weight (BW).

Figure 6

(a) Fourth-order polynomial curve fit of vertical ground reaction force data from Martin and Marsh (1992) and (b) typical experimental step force approximation compared to theoretical actual force profile (example for target 80 N load)

To impose the desired force profile on the transducers, the test stand shown in Figure 7(a) was constructed. The stand consists of a rigid aluminum frame, pneumatic piston, control valves, and measurement sensors. High pressure air (~110 psi building supply) was fed into the system which was then regulated to the desired pressure by either an electronic or manual pressure regulator. A three-way solenoid valve (Parker 71315SN) then applies the supply pressure to the piston or exhausts the piston pressure to the open atmosphere. Pressure inside the piston was monitored using a Wika model A-10 pressure transducer. The force exerted on the cymbal transducer was then inferred from the piston pressure using the piston geometry. For this work, a Bimba double acting piston of 2 in diameter was used.

Figure 7

Experimental methods, including (a) custom test stand, illustration of flow of information and testing of single transducer, (b) multiple transducer testing using custom test stand and simulated floor tile top, and (c) testing using human step inputs

In practice, the force profile of Figure 6(a) is difficult to be accurately recreated using the described test rig. The difficulty arises from the limitations of various commercially available electronic pressure regulators to regulate the pressure quickly enough or without unacceptable amount of overshoot. After several trials, the electronic pressure regulator was replaced with a manual regulator, which was set at a pressure that results in the desired force amplitude. The three-way solenoid valve then creates a pseudo-square-wave of force, which was taken to be an experimental approximation, and is shown in Figure 6(b). For this approximation, inlet and exhaust pneumatic piping size was chosen such that the rising and falling slopes of the approximated force profile most closely matches the theoretical gait force profile.

In Figure 7(a), force was applied to single transducers using a hard rubber (~75 A Shore hardness) tip to evenly distribute the load. Multiple cymbals were tested using the test stand, as in Figure 7(b), by applying the load through approximately rigid acrylic plates simulating a floor tile surface. Measurements were taken using National Instruments MyDAQ and NI9229 platforms. Finally, the system was tested in a floor tile enclosure under human footstep loads, as detailed in Figure 7(c).

PZT packaging

The brittle nature of lead zirconium titanate (PZT) can cause the formation of cracks over time when operating under cyclic loading conditions. Such cracks in the bulk material can reduce performance due to a complex combination of factors (Zhang and Gao 2004). This effect has been shown in practice by the East Japan Railway Company when testing their “Power-Generating Floor,” where they expressed the need to package the piezoelectric elements with a protective rubber structure (East Japan Railway Company 2008). It has been shown that PZT may be made substantially more compliant by creating a macrofiber composite (Wilkie et al. 2000). It is, however, simpler and cheaper if some of the benefits realized by a fiber composite could be realized using bulk piezoelectric material. Therefore, in this study it was examined if a bulk laminate composite can be used to improve the performance of the cymbal transducer and keep the cost of ceramic at the minimal level.

To minimize the onset of brittle fracture or cracking in the PZT, it should be laminated with a more ductile material. Additionally, the laminate should have a higher Young’s modulus than PZT in order to adequately transfer strain across any cracks which do form, but should be kept sufficiently thin so as to take a minimal amount of strain energy away from the PZT layer. It was thus decided to laminate the bulk PZT pieces with brass, because of its high ductility, electrical conductivity, and Young’s modulus approximately 40% greater than PZT.

To examine the influence of the laminate brass layer, three cases illustrated in Figure 8(a) were considered, including a baseline case of unlamented (plain) PZT, a 25 μm brass laminate case, and a 50 μm brass laminate case. The brass was laminated at the top and bottom surfaces of the PZT using All-Spec CW2400 conductive epoxy, bonded in a kiln at 80°C for no less than 2 hours. For these tests, smaller scale test specimens were used to minimize the material waste during ultimate breaking strength tests. The test specimens were of similar geometry to the dimensions shown in the drawing of Figure 4(a), however only 10 mm in depth. The test specimens were then tested individually using the pneumatic test stand, as shown in Figure 7(a), in terms of energy output and ultimate breaking strength. Energy output was determined by recording the voltage in a storage capacitor when the test specimen was subjected to a force impulse of 20 N load for 0.63 seconds. The storage capacitor was a 10 μF electrolytic capacitor and connected to the output from the test cymbal through a full-bridge diode (BAV21) rectifier. Energy produced was then inferred by measuring the voltage across the capacitor, by the relationship, 1/2CV2, where C is the capacitance and V is the voltage of the capacitor respectively. Ultimate breaking strength was determined by slowly increasing the load on the test cymbal until the moment of failure. Results are given as a percent increase or decrease over the unpackaged case, as it is a comparative analysis, and to avoid confusion with the reported performance of the full size cymbals.

Figure 8

(a) Illustration of three packaging techniques investigated including plain PZT, PZT with 25 μm brass laminate, and PZT with 50 μm brass laminate. (b) Results of packaging test, comparing performance in terms of ultimate breaking strength and energy output per simulated step input

As shown in Figure 8(b), the addition of the brass laminate causes a dramatic increase in breaking strength, as well as an increase in energy output per simulated step input. The increase in energy production is postulated to result from a more even stress distribution in the PZT caused by the addition of the laminate. The increase in breaking strength is more intuitively caused by the additional cross-sectional area and stiffness and subsequent reduction in stress per load. We see that while both laminate cases offer improvements in both categories, the greatest improvement, which we define as percent improvement in energy output per simulated step multiplied by percent improvement in ultimate breaking strength, is produced by the 25 μm laminate. Therefore, this packaging scheme was used for subsequent implementation in the tile.

Optimal number of transducers

We now describe the selection of optimum number of cymbal transducers that should be employed within the tile enclosure. The number of cymbal transducers determines the stress each one of them will experience and subsequently the charge produced by the PZT layer. The larger the number of transducers used, the more potential there is for energy to be harvested before the cymbals become overloaded. However, the more transducers which are used, the lower the sensitivity of the tile, or the amount of energy produced per load (i.e. person’s weight). Consequently, there exists an optimum number of transducers to be employed for maximum energy harvesting from a target weight. Figure 9 illustrates this concept by showing a qualitative comparison of energy produced per step on the tile, as a function of weight for an optimum number of transducers for the weight of an average child (8 year old) (Fryar, Gu, and Ogden 2012), average adult (Fryar, Gu, and Ogden 2012), and average NFL player (Merron 2004), as examples.

Figure 9

Qualitative illustration energy production per step as a function of weight for several cases of optimal number of cymbals used for target weights of a child (8 year old) (Fryar, Gu, and Ogden 2012), average adult (Fryar, Gu, and Ogden 2012), and average NFL player (Merron 2004) as examples

The flat areas of the curves in Figure 9 represent condition where the transducers would become overloaded and experience failure if not for the design of the tile enclosure (discussed in the Tile Enclosure Design section). We see that a tile optimized for a child can make much more energy per step at low weights, but has a much lower maximum possible energy production compared to a tile optimized for an adult or NFL player. We do note, however, that after a certain minimum weight, the floor tile will produce a constant amount of energy per step, regardless of the load. This type of performance is quite desirable as the function of tile (i.e. sending a signal when stepped upon) can be decoupled from a person’s weight or gait style. Therefore, we determined the optimal number of transducers such that the maximum possible energy harvested per step is just sufficient to power the signal transmission circuitry. For any person above a certain minimum weight, the tile will function in a similar manner. For reasons discussed in the Circuit Design section, this minimum weight was found to be around 36 kg and an optimal number of cymbal transducers was found to be five. In this application, a minimum necessary number of transducers were used. However, if the tile was built to harvest as much energy as possible, the number of cymbal transducers would be chosen to correlate with the expected average weight of occupants that would step on the tiles most regularly.

Tile enclosure design

The main functions of the tile enclosure are to provide a flat and rigid surface to transfer force to the cymbal transducers, protect the cymbals from overstraining, and protect the cymbals and circuitry components from the environment. To ensure a consistently flat and sufficiently rigid surface, it was chosen to construct the tile enclosure from Delrin^®, a thermoplastic which exhibits high stiffness, dimensional stability, low thermal expansion, and resistance to water and other chemicals. Channels were cut (CNC Mill) into the top and bottom plates, into which the supports are fitted. The support strips were bonded only to the bottom plate, using Loctite 120 HP epoxy, and were sized such that they do not come into contact with the top plate when the tile is not loaded and the top plate rests on the cymbal transducers. In Figure 10(a), we can see that when the tile is loaded, the cymbal transducers are compressed, although before the piezoelectric layer is strained to the point of failure, the top plate comes in contact with the supports, preventing further deflection. In this way, the cymbal transducers are protected from failure due to high loads. In addition, after a certain minimum load (i.e. person’s weight), energy production is held relatively constant, as the deflection of the cymbal transducers is mechanically limited. This leads to a more consistent operation (i.e. occupancy sensing and signal transmission), as a known amount of energy is delivered to the circuitry.

Figure 10

(a) Illustration of tile enclosure parts in unloaded and loaded states, (b) fabricated enclosure (with top plate removed) used when testing with human step inputs, (c) final fabricated enclosure (with top plate removed) with all components internal to the tile, (d) dimensioned drawing of floor tile enclosure, and (e) fully assembled STEP Tech floor tile with size 9 (US) shoe for scale

When developing the cymbal transducers and circuitry with human step loading, as shown in Figure 7(c), the output from the cymbal transducers was run outside the tile enclosure to the data acquisition (DAQ) modules, as shown in Figure 10(b), through small holes in the enclosure, which were later filled. Once the system was verified to function as anticipated, the circuitry was wired into the enclosure and superfluous wire which was used to reach the DAQ equipment in Figure 10(b) was wound in the extra space within the tile as shown in Figure 10(c). A dimensioned drawing of the tile enclosure is shown in Figure 10(d) with the fully assembled STEP Tech floor tile shown in Figure 10(e), along with a size 9 (US) shoe for scale.

Electrical connection strategy

While the cymbal transducers are loaded mechanically in parallel, their electrical connection strategy must be decided in order to realize the most efficient operation of the circuitry. When a 71 kg person stepped on the tile, it was seen that the output open circuit voltage of the cymbals is in range of 20 V to 30 V. Therefore, it can be expected that the maximum power point (MPP) will be one half of the open circuit voltage. If we consider connecting the cymbals in series, this would yield the maximum power point voltage of approximately 75 V in the case of a tile with 5 cymbals. This was derived under the assumption that all cymbals were the same, and that their maximum power point voltage was 15 V for each cymbal. In this case, the energy management stage would be faced with the task of converting the 75 V stored on the capacitor to 3 V, which is the voltage typically used by sensors and wireless modules. After searching for high efficiency DC/DC converters with the required input voltage rating, it was concluded that the conversion from 75 V to 3 V would be very inefficient. Furthermore, integrated solutions for low power switching DC/DC converters capable of performing the task are not easily available. Therefore, connecting the cymbals in series was not an option. On the other hand, connecting the cymbals in parallel would result in approximately 15 V maximum power point. At this voltage level, it was possible to use off-the-shelf converters, while providing high efficiency in energy transfer from the buffer capacitor to the rest of the circuitry. It was therefore decided to connect the cymbals electrically in parallel.

Rectification of the AC signal

In the literature, several different approaches have been demonstrated for piezoelectric output rectification (Beeby and White 2010; Erturk and Inman 2011; Guyomar et al. 2005). These circuits have been primarily designed to operate with stable periodic oscillations that would generate a predictable output of the harvester. However, the output of the cymbals when stepped on is an irregularly shaped single pulse. Furthermore, the amount of available energy after one step is in order of one millijoule. Hence, the rectification circuit selected has to be very efficient in order to not consume more energy than the gain of using the selected rectifier circuit. Furthermore, the storage capacitor can be considered empty before each step occurs, as an indeterminate amount of time will pass between the subsequent foot falls on the tile. Advanced rectifier designs typically require stable power for the control circuitry. All these limitations made the use of the advanced rectification a very challenging task. Therefore, a standard diode bridge was selected.

Selection of a storage capacitor

As with many energy harvesting applications, in this specific tile application there is not sufficient instantaneous energy produced to power the payload device (wireless signal transmitter in this case) continuously. Therefore, an intermittent storage capacitor is necessary to accumulate energy until signal transmission can be powered. Here, there is no need for accumulating energy between transmissions, principally because occupancy has already been counted, and also because energy is only collected when the tile is being stepped upon. Therefore, the capacitor buffer size was selected based on the value that would produce the most optimal energy output from a single step. Empirically, the value of the buffer capacitor was selected to be a 10 µF ceramic X7R capacitor rated for 25 V, as it stores the most energy per step, demonstrated in Figure 11. Note that in Figure 11 one simulated step input was signified by two rises in capacitor voltage, where one rise is caused by force being applied and the second rise by the force being removed. The regions of decreasing voltage in-between rises are caused by the high leakage current of the Schottky diodes used in the full bridge rectifier above 15 V. Schottky diodes are thus not used in the final circuit design.

Figure 11

Stored voltage for five cymbal transducers in parallel under 410 N simulated step for varying storage capacitor values. Values of maximum energy stored per step per cymbal per load are noted for each capacitance value

Energy management

Wireless signal transmission was performed using the ez460-RF2500 platform by Texas Instruments (Texas Instruments 2009). The energy required for starting up the transmitter and performing a wireless transmission, based on the documentation provided by Texas Instruments, was approximately 100 µJ (Morales and Shivers 2007). The amount of energy harvested during the stepping onto the tile, by an approximately 71 kg person, was borderline sufficient to perform the transmission. This, however, indicates that a lighter person wouldn’t generate sufficient energy, by stepping onto the tile, to perform a transmission. In order to address this issue transmission was delayed until the person steps off the tile. A similar amount of energy is generated as a person steps off the tile as when stepping on. Therefore, as is addressed in the Optimal Number of Transducers section, the minimum weight of a person necessary to power signal transmission is around 36 kg. Based on the measurements, when the stepping off is completed, the voltage reached values around 15 V. As a result, it was necessary to convert the voltage on the storage capacitor to the range of 1.8 V to 3.6 V, which is required for stable operation of the radio and MCU on the ez430-RF2500 board. The conversion is done by using a switching DC/DC regulator. Such a regulator is found in the LTC3129-1 nanowatt buck-boost DC/DC (Linear Technology Corporation 2013). This DC/DC converter was selected first because the circuit is rated for the voltage range expected on the buffer capacitor, and second it has analog control, which can be used for selecting a threshold voltage at which the converter should activate. By applying a voltage level above 1.22 V on the analog control pin, the converter would activate. Therefore, by providing the voltage from the input using a simple resistor divider it was possible to set the activation voltage at a desired input storage capacitor voltage level. However, this circuit has a narrow input hysteresis, so it is necessary to extend it in order to allow the output to use as much energy as possible before the DC/DC converter turns off. The hysteresis extension is implemented by connecting a diode and a capacitor to the RUN input, which controls the converter’s operation, as shown in Figure 12(a).

Figure 12

(a) STEP Tech circuitry schematic, with (b) qualitative illustration of circuit function, and (c) measured voltage response of circuit to simulated footstep forcing input to the tile

The circuit rectifies alternating voltage input from the cymbal transducers during a step. As the voltage on the input capacitor, C8, rises, the voltage on the capacitor connected to the RUN pin, C7, also rises, at a rate defined by the voltage divider and the diode voltage drop. When the voltage on capacitor C7 reaches the threshold voltage of the converter, 1.22 V, the converter starts operation. As soon as the converter starts operating, the voltage on capacitor C8 will drop, as the harvester is not capable of supplying sufficient power to prevent this, as the footstep on the tile is completed. With the drop of voltage on C8, if it weren’t for the diode D5, the voltage on the RUN pin would reflect the voltage change on the input, leading to disabling the converter once the voltage on this pin reaches 1.11 V. However, as the diode is preventing the discharge of the capacitor C7, the LTC3129-1 continues operation, draining the input capacitor until it reaches 1.9 V, where the LTC3129-1 enters under-voltage condition and stops powering the output. This process is qualitatively illustrated in Figure 12(b), with measured voltages on the input and output sides of the DC/DC converter under simulated stepping loads shown in Figure 12(c).

Introduction of the diode D5 and the capacitor C7 to the circuitry has the effect of realizing a low pass filter on the RUN pin. This in turn introduces a delay in response to the changing input. The value of the resistance divider is in megaohm range, in order to reduce the losses during charging. Therefore, the time constant of the RC network would be in the second range, depending on the values of the resistors and capacitor used. This delay could be potentially used to the advantage of the circuit. Let VTHR-DC/DC be the voltage level required for the input storage capacitor to reach in order to provide sufficient energy to perform a wireless transmission. This voltage level will be reached, as explained earlier, during the stepping off from the tile. If a person’s weight is the same as the minimum weight required for the system to operate, after the person has stepped off, the buffer would reach the required voltage VTHR-DC/DC after the delay introduced by the RC circuit. However, if a person is heavier, the voltage on the input buffer capacitor would continue to rise as the circuit is not being activated because of the said delay. In this way, more energy will be accumulated in the storage capacitor before the circuit is activated, hence providing additional energy to the circuit, increasing robustness.