Miniature water flow energy harvester based on savonius-type microturbine: an experimental study

In this experimental study, a miniature turbine-based water flow energy harvester designed for the purpose of providing power to wireless sensors within water pipes is reported. The device comprises a Savonius-type turbine and a radial flux permanent magnet electromagnetic generator. The two are magnetically coupled so that, while the turbine is submerged in the water flow, the generator operates in air. The device is cylindrical with a diameter of 0.8 cm and a length of 7.2 cm and, when inserted through a hole in a pipe wall so that only the turbine protrudes into the flow, it presents a cross-sectional area to the flow of only 1.25 cm2. Manufacturing was achieved through a blend of conventional machining methods, laser cutting, rapid prototyping, and the utilization of flexible printed circuit board technology for the generator stator. To ensure low friction and minimize cut in speed, ceramic ball bearings were employed. The prototype can function effectively at water velocities as low as 0.5 m s−1, generating electrical power within the range of 125 µW–5.1 mW when subjected to flow speeds between 0.5 and 2 m s−1. A maximum overall efficiency of 2.2% is achieved, when the water speed is 0.8 m s−1. Performance curves derived from experimental testing of the turbine for a range of rotor designs, obtained on a water flow rig, are presented and discussed.


Introduction
Nowadays, energy harvesting (EH) plays an increasingly vital role in powering autonomous microsystems.Among other methods of harnessing energy, EH from fluid flow has emerged as a potential solution for powering wireless sensors in pipeline networks [1][2][3][4][5], eliminating the need for batteries or costly and impractical wired monitoring schemes.Especially in water infrastructure, these sensors contribute to water quality monitoring [6] and structural health monitoring [7] by tracking quantities like temperature, pressure, pH and the concentrations of a variety of chemicals [8].
There are two prevalent methods for harvesting energy from fluid flows in pipes and ducts, one based on flow-induced vibrations [9], and the other on miniaturized turbines.Work on EH from water flow has been focused on the latter, and specifically on in-line devices that (a) require a break to be made in the pipe for installation and (b) typically present a significant obstruction to the flow.Installation of such devices is disruptive, requiring service interruption which is undesirable in a water network.Significant obstruction of the flow can also be problematic as it generates additional pressure drop.In the work reported here we aimed to address the question of whether it is feasible to implement an alternative type of water flow harvester, one that can be inserted through a small hole in the side of a water pipe, allowing installation with minimal disruption by hot-tapping (installation without service interruption) [10], compatibility with various pipe sizes, and a low degree of obstruction to the flow.Figure 1 illustrates the concept.
If the cross-sectional area of the device in figure 1 is small compared to that of the pipe, then it will behave in a similar fashion to a free-stream energy extraction device.The electrical output power that can be extracted by such a device may be expressed as: where ρ is the density of the fluid, V 0 is the free stream flow speed, A is the cross section presented to the flow, and C p and η are dimensionless parameters characterizing the power conversion efficiency.The parameter C p , known as the power coefficient, is the ratio of the raw mechanical power extracted to the power available P av in a cross-sectional area A normal to the flow: The parameter η is the mechanical to electrical power conversion efficiency.This can be further analysed as η = η m • η e where η m is the mechanical efficiency which accounts for the power lost due to friction and windage, and η e is the electrical efficiency which accounts for power lost due to the Joule heating in the generator stator coils.The product ηC p represents the ratio of the electrical output power to the available fluid power and is commonly referred to as the overall efficiency.
According to theory, C p cannot exceed a maximum theoretical value of 16/27 = 0.593, which is known as the Betz limit [11].Moreover, much lower C p values are expected at cm-scale [12] because of factors such as the relatively high viscous losses at low Reynolds numbers [13] and the inevitable increase in the relative sizes of clearances.However, while cm or mm-scale turbines cannot achieve performance levels near the Betz limit, they can still provide sufficient power for sensor applications, ranging from µW to a few mW.
To date, limited research has been conducted on miniature water flow turbine energy harvesters (WFTEHs).In one of the first attempts [14], aimed at powering flow meters in domestic water pipelines, a >30 mm (exact value not specified) diameter vertical axis propeller was combined with a 3-phase, radial flux generator.The resulting device required a minimum flow rate of 3 l min −1 to start and could produce up to 720 mW at a flow rate of 20 l min −1 .Other interesting works implementing a vertical axis propeller can be found in [15,16].A 20 mm-diameter horizontal axis turbine was implemented in [4].The harvester, also aimed at powering flow meters, could deliver up to 25.2 mW at 1 m s −1 flow when connected to a power management circuit (PMC).In [17] a device with a total outer diameter of 25.4 mm, employing 3D printed components is reported.The harvester can produce 0.5 mW at 0.25 m s −1 , and 4 mW at 0.46 m s −1 water speed.A multi-functional device (self-powered sensor for water flow rate and temperature) with a 4.8 cm diameter turbine is reported in [18].The harvester can generate 2.1 mW of power when the water speed is 1.75 m s −1 implementing a vertical axis propeller to harvest water entering from nozzles.An efficient 40 mm-diameter water flow energy harvester based on a horizontal axis propeller is reported in [19].The device, implementing a permanent magnet generator and designed to be employed in circular pipes with a diameter of 41.5 mm can generate 490 mW at a water speed of 1.85 m s −1 exhibiting a maximum overall efficiency of 5.15% when the water speed is 0.6 m s −1 .
Moreover, besides documented implementations, several companies have successfully commercialized WFTEHs.For instance, Kinetron, specializes in plug-and-play turbine-based harvesters, which are primarily in-line devices utilizing nozzles.Their products qualify as some of the smallest and highest power density commercial products available in this domain (most of the designed turbines have a diameter of ⩽20 mm) and are designed for applications including water heating, management, smart homes, and irrigation.
In contrast to the situation in water flow EH, numerous groups have demonstrated free-stream, turbine-based airflow harvesters which could provide inspiration for the development of WFTEHs.In [2], Howey et al produced a miniature shrouded wind turbine that could deliver between 80 µW and 2.5 mW of electrical power at air speeds in the range 3-7 m s −1 .The harvester had a rotor diameter of 2 cm, and a permanent magnet machine with an axial flux design was incorporated within the protective enclosure resulting in a 3.2 cm external diameter.More recently, a wind energy harvester with a 5 cm rotor diameter that could generate 40.7 mW at 15 m s −1 wind speed was presented by Fang et al [20], while Tomasini et al reported an airflow harvester with a 4 cmdiameter rotor for powering sensors used in railway vehicles [3].The latter device can reach a remarkably high net efficiency of about 15% when connected to a PMC.
Table 1 summarizes the state-of-the-art in cm-scale air/water turbine-based energy harvesters that employ turbines with a diameter ⩽5 cm.To esnure a fair comparison, the electrical power density values in the right-hand column are determined by considering the entire cross-sectional area exposed to the fluid flow.From the data in table 1 it is evident that cm-scale, in-line water flow energy harvesters are capable of generating useful levels of power (mW or greater) at realistic flow speeds (⩽a few m s −1 ).However, in-line devices are expected to be able to extract more power than free-stream ones because (a) they intercept the entire flow and (b) they can generate a larger differential pressure between input and output (within the limits set by the system).A key question to be answered in our research was, therefore, whether a cm-scale turbine of the type in figure 1, which is effectively a free-stream device, is also capable of generating useful levels of electrical output power.Another important question to be addressed was whether it would be possible to implement a generator with sufficient output voltage to make subsequent power conditioning manageable [21].This can be an issue in small devices at low flow rates because of the limited number of turns on the stator coils and the low rotation rate.
The WFTEH presented here is intended to form the power supply for a minimally invasive water pipeline sensor probe such as the one presented in [10].The idea behind such sensors is that they can be inserted through a hole in the pipeline wall that is sufficiently small so as to have negligible impact on the mechanical integrity of the pipe.With this potential final application in mind, the water turbine was designed for deployment in water pipes with minimum diameters of 100 mm, and for operation flow speeds up to 2 m s −1 .The following specifications derive directly from the application requirements: • Rotor diameter should be as small as possible to ensure insertion of the probe is minimally invasive.A diameter of 8 mm was chosen as the smallest size that was deemed practical with the planned fabrication route.• Rotor height should be ⩽25 mm so that additional height for various sensors is allowed while keeping obstruction of the harvester to the pipe to a minimum.
In sections 2 and 3 of this article we describe the design, fabrication and testing of a miniature WFTEH aimed at meeting the above requirements.A brief evaluation of the benefits and opportunities associated with the new device, along with a comparison with state of the art, are presented in section 4. The article concludes in section 5 with future considerations.

Overall device design and fabrication
Figure 2 shows exploded and cutaway views depicting the structure and construction of the device reported in this paper.The device comprises a 0.78 cm-diameter Savonius turbine coupled to a permanent magnet electromagnetic generator.The coupling between turbine and generator is magnetic and occurs across a window which separates the wet (turbine) and dry (generator) sections of the device.The overall cross-sectional area presented to the flow is only 1.25 cm 2 , which represents only 1.6% of the cross-section of a 100 mmdiameter pipe.To our knowledge the turbine is the smallest viable water-flow device reported to date.
It was decided that the generator should be subject to the same outside diameter constraint as the turbine, and consequently the generator had to be relatively long to achieve a reasonable volume and hence power generation capability.A radial flux (as opposed to axial flux) design [22] was chosen because it is more compatible with a long cylindrical geometry, and a 2-pole configuration based on a single diametrically-magnetized cylindrical magnet was adopted for ease of fabrication.A flexible PCB carrying the stator coils is glued to the inner walls of the upper casing.
Magnetic coupling is effected by two arrays of 2 mmdiameter disc magnets which are glued to end bushes on the generator and turbine sections of the device.These arrays sit either side of a polyimide window in the end of the generator section when the device is assembled.Magnetic coupling was used to ensure that as little of the device as possible is exposed to water so as to minimize corrosion and fouling.Also, increased viscous losses due to shearing of the fluid in the gap between the magnet and the stator would be expected if water were to enter the generator casing.An O-ring is used to set the clearance between the magnets on the turbine section and the window, and silicone sealant is applied around the coupling region to prevent leakage of water out of the turbine section.
The Savonius rotor is sliding fit on a shaft which is screwed into the bush at the end of the lower casing.Prototype turbine rotors were designed using SOLIDWORKS software and fabricated using a high-resolution 3D printing process (Stratasys Objet 500).Due to the ultra-small dimensions, a high-resolution rigid Vero White material was used for 3D printing.Vero White ensures high-quality rendering of the features at all scales (the smallest dimension was about 0.3 mm).3D printing could provide a viable manufacturing route for the turbine rotors in the longer term, but only if the printed parts can be protected from exposure to water, for example by metalisation.
Referring to figure 2, the bearing mounts, bushes, and shaft were designed with SOLIDWORKS software and fabricated in stainless steel by conventional machining.Their design was primarily aimed at achieving smooth operation and robustness for the device by eliminating, as far as possible, radial, and axial play of the shaft during rotation.Stainless steel was used to avoid issues with corrosion.The rotor is supported by radial full ceramic bearings (Boca BearingsInc.type R133-TP/C3 Z/S #5 AF2) with an inner diameter of 2.38 mm and an outer diameter of 4.762 mm, which were selected for low friction.The bearings are fitted into bearing mounts that are sliding fit in the casings.
The generator casing was fabricated from a length of stainless steel tube by laser cutting with a nanosecond pulsed ultraviolet laser (Spectra Physics type Talon 355-17).Serpentine channels were laser-machined into the tube to diminish the eddy currents generated in the stainless steel by the changing magnetic field.These channels were later filled with epoxy resin to improve the rigidity of the casing.The flexible PCB process (Stevenage Circuits Ltd UK) had minimum track and gap widths of 102 µm and the via pad diameter was set to 500 µm.Given these limitations, each spiral had the capacity to hold up to 23 turns at most.The nominal track height was 35 µm.A 2-layer stator was implemented, with coils that spiral inwards in one layer and outwards in the other layer.Figure 3 displays an image of the fully assembled harvester with the rotor mounted.Also visible is a Hall sensor that enables the generator to operate as a motor when needed.

Turbine design methodology and parameter selection
Several systematic reviews of Savonius turbine design and optimization can be found in the literature (see for example [23,24].The Savonius turbine is recommended over other counterparts like the Darrieus or H-rotor (Giromill) for low flow speed applications due to its superior self-starting capability, despite the trade-off of reduced efficiency.Savonius turbines are drag-based devices, meaning that they operate mainly due to drag forces acting on their buckets.In such devices, the peripheral speed of the rotor is lower than the speed of water flow.Consequently, lower values of C p can be expected typically falling between 0.05-0.30for larger scale turbines and even lower values being expected for miniaturized Savonius turbines [24].
Whereas lift-based horizontal axis turbines can be analyzed using blade element momentum theory [11], this relatively straightforward approach cannot be applied directly to Savonius turbines due to non-uniform drag distribution and mutual interference of the buckets.Computational fluid dynamics (CFD) simulations can be used in design, but experimental testing is the most reliable method for performance evaluation, and this is the approach taken in the present study.A review of Savonius turbine design at cm/mm scale was conducted to determine critical dimensions and geometrical parameters impacting efficiency and self-starting capabilities.Figure 4 shows the top view and a 3D model of a generic Savonius turbine rotor, along with the most crucial characteristic dimensions.
Given the restrictions imposed by the application, some geometrical characteristics were automatically pre-decided.For example, due to the small rotor size, it was considered essential (for robustness) to attach the rotor blades to a central hub, and this necessarily resulted in the Gap Ratio and Overlap Ratio both being set to zero in spite of this been nonoptimal according to earlier studies [25,26].Also, while it is generally recommended to have rotor end plates with diameters 10% larger than the rotor diameter [24], it was considered more important to maximize the rotor diameter, and since both diameters were subject to the same max constraint (must fit through 8 mm hole), both were set to 7.8 mm.As there was limited information to guide the rotor design at this scale, an experimental parametric study was carried out with a view to finding the optimum design.This involved fabricating and testing a range of rotor designs with variations in the Aspect Ratio(AR), Blade orientation and shape, Blade number and Number of stages.These parameters have been shown to influence performance in earlier work on Savonius turbine design.For example, it has been observed that increasing the AR typically improves performance [23,27], as does introducing twist into the blades to reduce torque fluctuations and enhance starting capabilities [28][29][30].
Regarding Blade number, according to [31], 2 bladed configurations show better power coefficient values while 3 bladed designs exhibit a decrease in the negative torque variations and hence higher torque coefficient values.These findings are in are consistent with the findings in [32].As a primary focus in this work was to enhance the operation of the device at low speeds, emphasis was put on the self-starting characteristics and hence a 3-bladed configuration was adopted.Finally, regarding the Number of stages, according to the literature, a two-stage turbine with a 90 • phase difference is the optimal arrangement from the multi-stacking configurations [33].However, rotors with two stages and a 90 • phase difference showed poor performance in early testing and were abandoned.

Generator design and modeling
A MATLAB code was developed to model and optimize the generator.Bearing in mind the challenges in efficient power conditioning for low-output voltage generators, emphasis was placed on maximizing the generator constant, defined here as the RMS (root-mean-square) open-circuit generator output voltage per unit rotation speed.The code initially computes the radial component of the magnetic flux density B r on a cylindrical surface surrounding a diametrically magnetized rod magnet using pole model analysis (PMA) [34].Subsequently, it constructs a spiral coil with vertices aligned to the same grid as the B r values and calculates the EMF In figure 5(a) the inner cylinder represents the permanent magnet with uniform diametrical magnetization M, while the stator coils, situated 0.5 mm away from the magnet, are illustrated as the curved surface of the outer cylinder (highlighted in red).The curved surface, S, of the magnet is divided into an array of surface elements: ds = n m ds, where ds is the area of the highlighted element and n m is the local unit normal vector.dl defines the length and direction of a short element of conductor within a stator coil, and n c is the unit normal vector at the location of this element.Then, r is the displacement vector joining the magnet and coil elements ds and dl.To calculate the magnetic flux density at the location of the line element some simplifications can be made to the general equation of the PMA : • since the flux density is being evaluated outside the magnet, M = 0 at the calculation point; • because the magnet is assumed to be uniformly magnetized throughout, the equivalent magnetic volume charge is zero; • consequently, just the integral over the magnet surface is left, where an equivalent magnetic surface charge ρ m = M • n m is assumed, with n m being a local normal unit vector.
The magnetic flux density at the location of the line element is then obtained as : Figure 5(b) shows a single layer stator with two identical spiral coils.The stator has been 'unwrapped', so the horizontal axis is the azimuthal position.For purposes of calculation, it was assumed that the stator is rotating about the z-axis at angular rate ω, and the magnet is stationary.The EMF contribution from the conductor element dl depends only on the radial component of B (i.e. the component that is normal to both the line element and velocity).This simplifies the flux density calculations as it becomes a scalar integral: Also, only the conductor elements aligned to the z-axis contribute to the EMF If the EMF is measured as V = V 2 − V 1 , then the contribution due to the highlighted conductor element will be dV = B r r c ωdl, and the total EMF will be: where the sign is + for elements on the right-hand side of the coil, and-for elements on the left-hand side.
Following the simulations, it was determined to employ a 6 mm-diameter, 14 mm-long diametrically magnetized NdFeB (neodymium-iron-boron) rod magnet (CERMAG grade N42).The outer dimension of the stator coil in the axial direction was set at 19 mm which was found to be optimal in terms of power generation.In general the coil should extend beyond the ends of the magnet so that all of the magnet's length contributes to the induced EMF; however, as the coil is made longer the generated power passes through a maximum, beyond which the increasing winding resistance becomes dominant.Assuming a remanent flux density of B T = 1.3 T for the magnet (as quoted by CERMAG), the generator constant for a single-layer spiral coil was estimated to be ∼59 µVrms/RPM, while the stator resistance per spiral was estimated to be 4.8 Ω.
Based on these values, the average power into a matched load was calculated to be ∼4.5 mW per spiral and hence ∼18 mW in total for the entire 2-layer stator, when the magnet is rotating with 5000 RPM.The Cu track width of the coils was assumed to be half the coil pitch (100 µm).The radial magnetic flux density distribution as derived from MATLAB code can be seen in figure 6(a).
Also, in figure 6(b), the simulated output voltage waveform generated by a single spiral when the magnet is rotating at 5000 RPM is shown.

Experimental setup
To examine the performance of the prototype harvester, a flow rig was used.The experimental setup is shown in figure 7. The flow rig comprises two test sections featuring distinct diameters; the smaller of the two, with a 28 mm internal diameter, was utilized in this work.Both test sections were designed in adherence to ISO 5167 [35], a standard that stipulates the necessity for minimum lengths of upstream and downstream piping before and after pressure differential devices.Water is pumped from a tank at the base of the rig using the CRESTPUMP AM -50 water pump controlled by a Parker AC10 motor controller.An ABB FEP611 electromagnetic flow meter with a measuring accuracy of 0.4% of rate (of measured value) is used as the reference volumetric flow meter.The water flow speed is controlled via an analogue input on the AC10 controller.The maximum achievable water flow speed is 2 m s −1 when 5 Volts are supplied.The WFTEH is mounted for testing in the middle of the pipe.It is inserted through an 8 mm-diameter hole in the pipe wall and sealed by silicon rubber sealant.Testing of each rotor design was carried over a range of fixed flow speeds.At each flow speed the electrical load applied to the generator was varied using a custom programmable load circuit, and the corresponding variations in the rotation speed, output current, output voltage and output power of the WFTEH were recorded.In each experiment the applied load resistance was stepped through the same set 24 values ranging from 500 Ω down to 10 Ω; an 'open-circuit' measurement was also made where the only load imposed on the generator was that required to measure its output voltage.In all the experiments the rotation speed, and hence the output power of the turbine, was limited ultimately by the maximum water flow speed the flow rig could provide, which was 2 m s −1 .Setting of the load and data acquisition were carried out with the aid of a custom LabVIEW interface.The variable load circuit could be re-configured to operate the generator as a motor at flow speeds where the turbine would not self-start.

Comparison of different rotor designs
Five different rotors, covering 3 different degrees of blade twist and 2 different aspect ratios, were designed (see table 2).The fabricated rotors can be seen in figure 8.After the evaluation, the one that showed the best overall performance, taking  into account both power generation performance and starting capability, was chosen for further analysis.
First the three rotors with identical aspect ratios of 2.05 (rotors I,II,III) were tested.Figure 9 illustrates the recorded variations with rotation speed of the power into the load when the blades were twisted by 90 • , 60 • and 180 • degrees respectively.Rotors I and III exhibit better overall performance in terms of power extracted and minimum required water flow speed to operate.Rotor II could not sustain rotation for flow speeds below 1 m s −1 .Also, for lower speeds Rotor I exhibits better behavior than rotor III and can sustain rotation when the speed is as low as 0.5 m s −1 .For this reason, rotors with 90 • and 180 • twist were selected for further testing while rotors with 60 • degrees twist were excluded from the optimization process.
After the effect of the twist of the blades had been investigated, the effect of AR was tested.Rotor I, together with 2 new rotors (Rotors IV and V) with increased ARs of 2.692 were evaluated.The results, shown in figure 10, confirmed that an increase in AR leads to an overall improvement in performance.For instance, comparing Rotors V and III, both of which have their blades twisted by 180 • , Rotor V consistently outperforms Rotor III in power generation across all tested flow  speeds.Also, all the rotors were able to sustain rotation when the fluid speed was 0.5 m s −1 , whereas in the previous tests only Rotor I could do this.However, Rotors IV and V needed assistance from the motor to start on two occasions, and this led to their exclusion from further testing.In the end, Rotor I was chosen for further investigation as it showed the best performance overall.

Rotor I performance
Figure 11 shows, for Rotor I only, the recorded variation of electrical output power with load resistance at different flow rig water speeds in the range 0.5-2 m s −1 .Below 0.5 m s −1 flow rate the turbine would stall, needing the motor to re-start, only to stop after a few seconds, so no trustworthy measurements could be conducted.Each data point shows the average The solid lines in figure 11 are fitted rational polynomial curves obtained using the rat23 function in Matlab.In all of the data sets, the load power passes through a single maximum, as expected for a voltage source with a finite source resistance.Based on the fitted curves, maximum power was obtained with a load resistance in the range 43-51 Ω. Figure 12 shows the load power recorded at each flow speed when the applied load resistance was closest to the optimum value.The actual applied load resistances are also plotted.
The maximum power levels recorded for Rotor I, as plotted in figure 12 were 125 µW at 0.5 m s −1 flow, 1.25 mW at 1.0 m s −1 , 3.24 mW at 1.5 m s −1 and 5.1 mW at 2.0 m s −1 .Assuming the load resistances in figure 12, the corresponding RMS load voltages are 79 mV, 237 mV, 382 mV and 462 mV.
Based on measurements of the open-circuit generator voltage, the generator constant per spiral coil was found to be 52 µVrms/RPM a result that closely aligns with the anticipated value.The small discrepancy could be due to the generator magnet having a lower remanent flux density than expected and/or the gap between the magnet and the coil being slightly larger than in the model.The stator coil resistance was also measured and found to be 41 Ω when all four spiral coils were connected in series.This is higher than the predicted value, the discrepancy presumably being due to the copper tracks in the PCB being thinner/narrower; the model also neglected the additional resistance due to PCB via holes.
Figure 12 also shows the variation with flow speed of overall efficiency, (ηC p ).To evaluate this parameter, the maximum load power was divided by the power available in the flow, i.e. equation ( 1) was rearranged to give: ηC p = P out / 1  2 ρAV 3 0 .A maximum overall efficiency of 2.2% was observed at 0.8 m s −1 flow.The overall efficiency dropped gradually with increasing flow rate above 0.8 m s −1 , reaching 1.0% at 2.0 m s −1 flow.A drop in efficiency was also seen at flow rates below 0.8 m s −1 .
Finally, figure 13 shows the variation with flow speed of the RMS generator output voltage for both the optimally loaded and open-circuit conditions.

Discussion
The initial results presented in the previous section are encouraging; they demonstrate that a viable water flow energy harvester along the lines envisaged can indeed be realized.The prototype device can generate ⩾1 mW at flow rates above ∼0.9m s −1 and this level of power is sufficient to support numerous types of wireless sensor node operating at low duty cycle and requiring only short-range communication [5].The output voltage levels are compatible with modern power conditioners employing active rectifiers (see for example [36,37]).However, the voltages at the lower end of the flow rate range are too low to support cold-starting, where the minimum required voltage for state-of-the-art rectifying power conditioners is around 250 mV [38].From figure 13 it can be seen that the open-circuit generator output reaches this level at a flow speed of ∼0.7 m s −1 .The device as currently implemented would, therefore, not be suitable for a battery-less sensor node if cold-starting at the lowest speeds was a requirement.
Referring to table 1, the turbine implemented in this work has a diameter 2.56 times smaller than the ones in [2,4] and 5.1 times smaller than the ones in [1,3,19].Moreover, it presents a cross-section of only 1.25 cm 2 to the flow and this is significantly smaller than the cross-section of any other device reported in literature to our knowledge.Small size is advantageous as it minimizes the perturbation of the flow due to insertion of the harvester.
When comparing the power generating capability of our device with previously reported water flow harvesters, it is necessary to bear in mind that in-line devices are not subject to the same constraints as free stream ones.In-line devices can operate at higher pressure drop for a given flow rate and as a result can achieve higher power densities.Referring again to table 1, our device is outperformed in terms of power density by [4,17,19].A fairer comparison can be made based on efficiency, but this is not always possible based on the data provided; this is the case for [4,17].For the device in [19], the maximum efficiency was 5.15% compared to 2.2% for our device.A comparison can be drawn with air-flow harvesters which are mostly free-stream devices, but then it is necessary to correct for the vastly different fluid properties.For example, if the device in [2] were to achieve the same overall efficiency when operated with water (∼830 × denser) at 2 m s −1 flow, it would achieve a power density of 0.61 mW/sq.cmwhich is comparable to the value achieved by our device at the same flow speed.
One notable feature of the experimental results is that the effect of electrical loading on the turbine rotation speed is relatively weak.This indicates that the generator is under-rated compared to the turbine and that more power could be extracted from the turbine by loading it more heavily.This observation is consistent with the fact that the optimum load resistance observed in the performance tests was only marginally larger than the stator coil resistance; if the generator were able to load the turbine more heavily, it would lead to a more significant drop in turbine speed, and this in turn would lead to an increase in the effective source resistance of the harvester and hence the optimum load resistance value.This problem is exacerbated at higher turbine output power levels, and this explains the downward trend in overall efficiency as the flow rate increases.An important difference between water flow and air flow energy harvesters is that the available power in a water flow is several orders of magnitude higher for a given flow speed due to the higher density.Consequently, it is less critical to maximize the turbine efficiency in the case of water flow.Instead, priority should be given to improving the matching between the generator and the turbine so that more of the available turbine output power can be extracted.For pipeline monitoring applications, where there is generally space available outside the pipe, an obvious way forward is to have the generator outside the pipe and increase its size since the power rating of permanent magnet generators scales steeply with size [39].For applications where this is not possible, improved generator performance should be achievable by increasing the fill factor of the stator coils and careful design of the magnetic circuit.
The prototype harvester has been operated for several days without any noticeable change in performance.However, the long term reliability of the device has not been tested.One potential issue with the current design is that the side-loading of the bearings in the turbine section is increased by having both bearings at the same end of the rotor.This was done to maximize the rotor diameter.Designs with a smaller-diameter, partially occluded rotor that allows a bearing to be supported at the distal end of the device will be explored.

Conclusions
The device presented in this paper is, to the authors' knowledge, the smallest turbine-based energy harvester reported to date.Aimed specifically at water pipeline monitoring applications, it can operate at flow speeds down to 0.5 m s −1 and deliver between 125 µW and 5.1 mW of electrical power at flow speeds in the range 0.5-2 m s −1 respectively, which are quite common for a water pipeline network.The overall efficiency of the device reaches its peak value (2.2%) when the water speed is 0.8 m s −1 .Flow rig experiments tests have shown that the generator is under-rated compared to the turbine, and consequently we are not able to run the turbine at its optimum operating point.Future work will focus on improvement of the generator design so it can load the turbine more heavily.Further work concerning optimizations in terms of repeatability and ensuring unattended long term operation as well as the development of a fully autonomous sensor probe will also be carried out.

Figure 1 .
Figure 1.Water EH concept explored in this work.The harvested energy may be used to power autonomous condition monitoring sensors in a wireless sensor network (WSN).

Figure 2 .
Figure 2. Technical drawings of the entire device showing (a) exploded and (b) cut-away views.

Figure 3 .
Figure 3. Assembled water harvester with Savonius-type micro turbine mounted, next to £1 coin for scale.

Figure 4 .
Figure 4. Key geometrical parameters of a Savonius turbine.

Figure 5 .
Figure 5. Generator model, showing (a) discretization of magnet surface (inner cylinder) and location of stator PCB (outer cylinder, in red), and (b) stator coils 'unwrapped' so the horizontal axis is azimuthal position.

Figure
Figure MATLAB simulation results: (a) 3D surface plot of radial magnetic flux density distribution on a cylindrical surface outside the diametrically magnetized magnet, (b) angular variation of output voltage generated by a single spiral coil at 5000 RPM.

Figure 7 .
Figure 7. Water flow test rig during testing.Inset image shows voltage scope displaying WFTEH output voltage.

Figure 8 .
Figure 8. Fabricated 3D printed rotors for the experimental study.

Figure 9 .
Figure 9. Measured variations of load power with rotation speed at various water speeds for 3 rotors with aspect ratios of 2.05 and twist angles of 90 • (I), 60 • (II) and 180 • (III).

Figure 10 .
Figure 10.Measured variations of load power with rotation speed at various water speeds for 3 rotors with aspect ratios of 2.05 (I) and 2.69 (IV,V).Twist angles are 90 • (I, IV) and 180 • (V).

Figure 11 .
Figure 11.Measured variations of load power with load resistance at different water speeds for Rotor I. Open circles are measured data; solid lines are curve-fits to the measured data.

Figure 12 .
Figure 12.Maximum load power (solid blue line), optimum load resistance (solid red line) and overall efficiency (dashed blue line) for Rotor I as a function of flow speed.

Figure 13 .
Figure 13.RMS generator output voltages under optimally loaded and open-circuit conditions as a function of flow speed.

Table 1 .
State-of-the-art cm-scale air/water turbine-based energy harvesters.
a Calculated from the paper.b Free Stream.

Table 2 .
Geometrical properties of the designed rotors.