GSM-enabled remote monitoring of rural handpumps: a proof-of-concept study

The continued expansion of mobile network coverage in rural Africa provides an opportunity for simple and low-cost hydroinformatic innovations to measure and transmit data on handpump use for policy and management improvements. We design, build and test a Waterpoint Data Transmitter to determine its robustness, functionality and scalability. Results demonstrate that this novel application using simple microprocessor, accelerometer and global system for mobile communications (GSM) components has significant potential in recording graduated time-step information flows of lever pumps which can be modelled into a reasonable water volume use approximation. Given the systemic informational deficit for rural waterpoints in Africa, where one in three handpumps is likely to be non-functioning, this innovation has the potential to provide universal, low-cost and immediate data to guide timely maintenance responses and planning decisions, as well as drive greater accountability and transparency in donor and government behaviour. doi: 10.2166/hydro.2012.183 s://iwaponline.com/jh/article-pdf/14/4/829/386770/829.pdf Patrick Thomson (corresponding author) Rob Hope Tim Foster School of Geography and Environment, University of Oxford, South Parks Road, Oxford OX1 3QY, UK E-mail: patrick.thomson@ouce.ox.ac.uk


INTRODUCTION
Handpumps have been a key technology for accessing groundwater for decades. Given the importance of groundwater as a safe and reliable source of water for the world's rural poor (MacDonald & Calow ) significant levels of effort and investment have gone into understanding and improving handpump technology (Arlosoroff et al. ).
However, despite their relative simplicity, the sustainable operation and maintenance of these pumps is an enduring challenge. It is estimated that across Africa around one in three handpumps are non-functioning at any given moment (RWSN ). However, there is lack of reliable and up-to-date information about rural water access (Jiménez & Pérez-Foguet ), and without widely available information on the status of pumps, operations and maintenance (O&M) is invariably conducted by local communities who face challenges with achieving economies of scale and a lack of technical and managerial expertise (Harvey & Reed ; Carter ).
Mobile phone coverage is now reaching rural areas that have to date enjoyed few other services (e.g. grid electricity or piped water supply). It is estimated that in 2012 more people in sub-Saharan Africa will have access to the mobile phone network than have access to improved water supplies (Hope et al. ). In many low to middle income countries, mobile voice and data networks are a key transforming technology, as they leapfrog past the traditional landline networks enjoyed elsewhere. There is innovation at all levels, from mobile phone charging with bicycle dynamos where there is no electricity supply, to mobile banking and payment services which has the potential to reach a vast segment of the population currently not served by traditional banks. Hope et al. (, p. 10) identify the increased mobile phone network penetration into rural areas as providing a 'platform for innovative technical, financial, and institutional solutions' to enable new management models for rural water services and enable effective regulatory oversight, through the provision of accurate and water resources management or assist in flood early warnings. However, it has not been used before for monitoring handpump usage. This is partly due to there not having been a perceived need for metering in this context and partly due to the unsuitability of existing technology.
Direct flow measurement devices are not designed to operate with handpumps, whose output is of low and varying pressure, through a relatively wide aperture. Attaching such a device directly to a handpump outlet would restrict the output, and thus raises issues for use-acceptability.
Additionally, existing smart meters are relatively high cost in comparison to the cost of a typical handpump. The increasing global system for mobile communications (GSM) coverage of rural communities, along with the ever decreasing cost of electronic components, increasing energy efficiency and improving battery technology, also driven by the mobile phone sector, enables the use of telemetry to be investigated in this context. This paper reports on the design and testing of a new Waterpoint Data Transmitter (WDT) that can provide reliable real-time data on handpump usage to address this information deficit. Using a low cost integrated-circuit (IC) based accelerometer, the WDT automatically monitors the number of strokes made in operating a handpump, and then transmits this information over the GSM network.
This provides volumetric output estimates that can show daily to seasonal demand levels, including critical underor over-usage information to inform repairs or to justify further investments. Information on rural water usage patterns can inform water supply infrastructure planning. The initial trials of the device, outlined in this paper, were conducted in Lusaka, Zambia, in July 2011 on India Mk.2 handpumps. The aim was to demonstrate proof-of-concept in a real environment, rather than conduct rigorous trials in a controlled laboratory environment.

Waterpoint Data Transmitter
The WDT, the prototype of which is discussed in this paper, aims to be a robust, low-cost and scalable technology, and all design decisions were made with these three characteristics in mind. The WDT is attached to the handle of a handpump and consists of three essential elements: (a) an IC-based accelerometer; (b) a microprocessor; (c) a GSM modem. Elements were chosen on a criteria of ease of use for prototyping rather than optimal performance.
The use of an accelerometer to measure handle movement was chosen for a number of reasons. First, given the number of handpumps in use across the world, a design that could be easily retrofitted to existing in-service pumps and that (other than adding insignificant extra weight to the handle) did not interfere with the pump in any way was essential. This also has advantages at the prototyping and test stages. Second, by using a solid-state IC there are no moving parts, which should increase reliability over other options (e.g. direct water flow or handle movement measurement).
The accelerometer used is similar to those found in high-end mobile devices or certain games console controllers. In this case an Analog Devices ADXL335 was used.
The accelerometer senses movement in X, Y and Z planes and produces three analogue outputs proportional to the acceleration sensed along that axis. The ADXL is capable of sensing ±3 g which was deemed to be sufficient for the purposes of the WDT. The analogue output from the ADXL335 was filtered by a simple resistor-capacitor (RC) filter implemented in hardware to reduce the bandwidth of the signal reaching the microprocessor. This was to reduce any high frequency acceleration noise that could confound the tilt calculation while not being relevant to the measurement of pump handle movement. The À3dB frequency of this filter was 2.17 Hz, which is approximately four times the observed fundamental.
The microprocessor for signal processing and control was an ARM mbed (www.mbed.org). This is a generic prototyping platform and was chosen for ease of use during prototyping and field testing. It should be noted that this unit is significantly over-specified for this application. The processor takes the acceleration data from the accelerometer and calculates a pump handle tilt angle. This tilt angle is then monitored to produce a count of the number of times the pump handle has moved over a given time period and an estimate of the volume of water abstracted.
This information is periodically sent as a short message service (SMS) message via the GSM modem. The GSM modem accepts standard AT commands from the microprocessor to send out periodic SMS messages. The test setup had the WDT sending messages to a recipient mobile phone once per minute. A full implementation of this system would involve a data terminal with a bespoke user interface and could use either SMS or global packet radio service (GPRS) data transmission protocols, sending data at the rate deemed most effective in terms of the trade-off between currency of data and cost of transmission.

Experimental setup and initial testing
Initial testing to characterise the pump and develop the algorithm was conducted on a pump owned by Geotech Ltd in Lusaka, Zambia, to which we were given unrestricted access. The WDT was strapped onto a pump handle towards the fulcrum to reduce distortion and to keep it from interfering with the users. The accelerometer within the WDT was approximately 30 cm from the fulcrum. To aid analysis a Nintendo Wii Remote (a.k.a. Wiimote) was also attached to the pump handle along with the WDT prototype (see Figure 1). The WDT sent periodic data to a mobile phone via SMS and the Wiimote sent real-time acceleration data via a Bluetooth link to a nearby PC. The raw acceleration data from the Wiimote were captured using freeware called g-force analyser.
The basis of the WDT algorithm is that the tilt angle of a body with respect to the Earth's surface can be derived from the ratio of the acceleration component on each axis due to gravity according to the simple equation (illustrated in Figure 2): This is only valid for an object at rest. When the object is moving, there are other acceleration components that act on the body and confound the calculation of tilt (see Figure 3).
In the case at hand, where the accelerometer mounted with its x-axis along the pump handle and its y-axis along the axis of handle rotation the acceleration components would be: (The sign change arises due to the frame of reference of the inner workings of the accelerometer.)     • number of strokes made (observed); • time taken to abstract 20 l (observed); • pumping characteristic, e.g. fast, slow, full stroke, half stroke (observed); • WDT stroke count (mobile phone); • raw acceleration data (Wiimote/PC).
The raw acceleration data captured by the Wiimote were analysed to see how the calculated tilt varied during pumping. Figure 5 shows raw data from a typical 20 l run calculated using Equation (1)  The WDT would then transmit the four scores from each window, allowing weightings for calculating volume to be made at the data centre. The windows chosen were: These windows were biased below the horizontal to reflect the observed pumping action, with the very high apparent excursions above the horizontal discounted on the grounds that they were quite dependent on pumping technique and movement here did not contribute to water output. They were overlapping due to the fact that to 'score' the pump handle would have to pass through the entire range of a window, not just into or out of a window.
For example, if the pump handle moved consistently between À25 W and þ5 W , a range that would produce a significant output of water, only the WDT would only register a score for W 3 and not for either of W 2 or W 4 . While these windows did not cover all possible stroke ranges they did cover all likely stroke ranges observed under normal use. Another 14 test buckets of 20 l were filled, with the WDT transmitting the four counts from each run to a mobile phone. These were then analysed in order to derive the relative weightings for each window.
The weights were calculated using MS Excel's Solver function. An additional simplified weighting was calculated.
The results are given in Table 1 For the live testing the WDT data (four 'window' scores and volume estimate) were recorded automatically as they were received on a mobile phone and the Wiimote recorded the raw acceleration data. Figure 7 shows both the Wiimote and WDT attached to the pump. The timings for each 20 l bucket and the person who had pumped were recorded by hand. A few problems were encountered with the electronic data capture, but none that caused substantive difficulties.
The mobile network occasionally dropped out, causing some accumulation of the SMS data and for SMSs to arrive in the wrong order, both issues that could be resolved at the data analysis stage. The link between the Wiimote and the PC dropped out from time to time, resulting in gaps in the raw data stream, but as this was a secondary data source, not critical to the study, this was not a problem.
Testing took place with the support of the Community Water Committee and with the consent of the pump users.
In order to generate data that were as representative of normal use as possible we did not try to influence who used the pump. The only request made was that users abstracted water in discrete 20 l volumes rather than mixing different container sizes. Initially, our presence generated a certain level of interest, and it became obvious that a handful of the pump users (men and teenagers) rarely used handpumps, if at all. However, initial crowds soon dissipated and our presence seemed to be of less interest.
By chance the second pump/borehole tested was faulty, with much greater effort required to pump out a given volume of water than usual. The two sections of the borehole casing had become detached so water was leaking back into the aquifer above the pump cylinder. However, during the time we were at Valley View this was repaired, enabling us to return for a before and after comparison.

RESULTS AND ANALYSIS
Data were collected over 4 days for three pumps as summarised in  Figures 8 and 9 show the data for pumps #1 and #3, respectively. These pumps were in good working order with no observable difference between them and the pump used for the initial testing and      algorithm generation, other than more wear and tear as they served a larger community. The WDT calculation tracks the observed output well over time, with the cumulative percentage error over the test runs being under 10%.
To a certain extent this is down to the luck of when the test was stopped but an accuracy of the order of 10% is accep- were by men, and the period from 20 to 30 min the pump was being used by girls. Over time, on the assumption that the WDT weightings have been generated from a representative range of users, these variations will average out: an operational system would be more likely to report data every hour rather than every minute, so these subtleties/errors would not be captured.
The data from pump #2 tell a slightly different story.
Observation of pump #2 on 13/7/11 suggested a serious mechanical fault. A lot more effort was required by users to abstract water and a period with no pumping during the changeover between users of around 10 seconds would require around 20 strokes of pump priming before water was flowing again. This was in contrast to pump #1 where such priming required only one or two strokes. Figure 10 shows the difference between the WDT calculation and What these differences do show is that to produce accurate data on pumped water volumes with the WDT, each pump/borehole combination needs to be characterised to generate the correct weightings. Depending on usage patterns and static level, the initial priming of the pump after a long break may be a greater or lesser issue. In the case of the pumps on which these tests were conducted, with their high usage levels, the number of times the pump will have to be primed each day in proportion to its total use will be quite low. For a pump supporting a smaller population with a deeper static level, priming will be a significant factor. Common to all pumps, however, will be the issue of leakage due to worn parts or poorly aligned casings.
The original thinking behind the WDT was that nonfunctioning pumps could be identified when their usage dropped from a normal level to near zero. What the problem encountered with pump #2 illustrates is that it may also be possible to identify another pump failure mode. If the apparent volume of water starts to deviate significantly from previously observed patterns or levels, this may indicate an issue with the borehole casing or another issue that similarly affects pump performance. A slowly increasing apparent volume may also indicate a problem with the pump. There are numerous exogenous factors that could also result in such an increase, for example an increasing number of people using the pump in question due to the failure of another water source, or a dry spell that requires more abstraction for non-domestic purposes. Looking only at data from one unit in isolation it would be very difficult to determine the exact cause, but with scale and time, temporal and longitudinal comparisons could be made that will reveal more.

APPLICATION AND IMPACT
The shift towards community participation and management in rural water supply from the late 1970s onwards was part of a wider demand-responsive approach (DRA) promoted by the World Bank in response to the weak sustainability of government water programmes (Kleemeier  In response to these challenges there is increasing interest in investigating professionally-oriented models for water service provision (Harvey ; Kleemeier ). There is potential for more clearly defined O&M responsibilities and potential for economies of scale and risk pooling.
Harvey & Reed () also suggest that private involvement is service provision as well as parts supply will allow for greater revenue generation and thus be a more sustainable business than spare parts provision alone. Carter () argues that rural water services should more closely resemble urban services which are run professionally with a larger user base over which to spread cost and risk. He states that there is a need for 'rural utilities' or service providers with a strong customer and performance-orientation, with local government taking a regulator role.
Reliable and frequent flows of information are required for both the efficient running of a professional O&M regime and its effective oversight by a regulatory body. A private maintenance contractor who is maintaining a portfolio of handpumps in a number of villages may achieve the economies of scale that Harvey () promotes, but will not be co-located with the handpumps and users it serves. Without information on handpump performance and failure a reliable, timely and efficient system to repair and maintain pumps is impossible. If no information is available, scheduled maintenance is possible but timely response to breakdowns is not. If repairs are only undertaken when faults are reported by users or village water committees, the system has a chance of working, but only in the case of a well-organised water committee and a motivated contractor. The WDT described in this article can provide this information. As well as providing timely and unambiguous data on the non-functioning of a pump which can be swiftly acted upon, analysis of recent historical usage data may provide some indication of the nature of the failure and thus speed up the repair cycle. Such transparent performance reporting will hold maintenance providers to account, and develop performance benchmarks that are currently absent from the rural sector. There are equivalent benefits in the case where rural water service provision continues to be implemented by government. The government ministry or department responsible will have the ability to monitor the effectiveness of delivery by local government agencies, with the potential to benchmark performance between regions or districts.
The WDT will also provide a rich source of data on historical usage levels and patterns. This can inform future investment and resource planning decisions as it will be possible to differentiate between heavily used and lightly used pumps, between those which break regularly and those which are more reliable. Limited resources can then be better targeted at those areas most in need, instead of relying on proxy measures for use such as population size, or having decisions heavily influenced by local politics.

CONCLUSIONS
The results of this trial indicate that the real-time monitoring of rural handpump functionality is possible. This has the potential to bridge the information gap that currently hinders efforts to efficiently maintain rural handpump networks, and enable new O&M models to be investigated.
A unit based around low-cost IC-based accelerometers and off-the-shelf GSM technology can monitor handpump usage and produce a useful, if imperfect, estimate of water abstraction. Four main challenges remain, which must be addressed before any comment about the larger-scale viability of the concept can be made.
First, the prototype was tested on only one type of pump, and the test pumps were all of similar age/condition and in a similar environment. Given the time constraints of this study this was the best approach. The system has been designed to be compatible with any lever-action handpump, with only offline calibration required for use with different pump types. However, further tests must be conducted on a wider range of pumps and boreholes to confirm that this is the case in practice.
Second, the issues for pump-priming and leakage will confound the volume calculation algorithm under its current implementation. (Simple spectral analysis of the signal produced when the pump is being primed and when it is producing water did not reveal any noticeable differences.) While requiring a more invasive approach to the pump, the integration of a water flow or temperature sensor would help resolve this problem, and provide the additional possibility of being able to monitor changes in the static level of the borehole in question.
Third, the battery life of the prototype was acceptable for these tests (of the order of a day). It used a highly over-specified processor whose current consumption was correspondingly high, and transmitted data more often than a production unit would be expected to. Nonetheless, power consumption remains the biggest technical challenge to be overcome before a viable production unit is possible.
The options of solar power and using the kinetic energy of the handle movement should be investigated.
Finally, environmental durability and user acceptability must be considered. Any operational implementation has to acknowledge the possibility of children playing with footballs near water points and other day-to-day hazards. While a new iteration of the design that can be fitted within the pump is being developed, this is not simply an issue of product design. An item such as this is at risk of vandalism and theft if long-term user acceptability is not ensured in the design and implementation of the service delivery model.
The residents of Valley View were gracious hosts and happy to assist in this study; however, no assessment was