A LEAK DETECTION SYSTEM

FIELD OF THE INVENTION

[0001 ] This invention relates to sensor synchronisation ideally suited to a leak detection system. In particular, a preferred aspect of the invention relates to synchronising a plurality of sensors in a leak detection system configured to detect leaks in a water distribution network.

BACKGROUND OF THE INVENTION

[0002] In many industries, it is necessary to synchronise sensor measurements from a plurality of sources. This is particularly the case in position determination methods that rely upon timing. However, even with modem technology synchronising discretely located devices to a high level of precision is challenging as many factors introduce delays and lag. One such application in which this is pertinent is in leak detection systems.

[0003] Fluid loss due to leaking pipes is an issue that continues to plague fluid distribution networks. Water pipelines, for example, are located underground and are not visible. Accordingly, when a leak occurs it often goes unnoticed and unrepaired. In this instance, water is continuously lost from the water distribution network. These types of leaks will often grow over time causing more water loss, and ultimately end up causing a burst pipe. Large leaks and water bursts are expensive, not only due to water loss, but due to repair costs and associated property damage.

[0004] Leak detection techniques exist but a number of issues exist with current technologies. No current technology can continuously and autonomously monitor a water network, for example, and pinpoint the location of a leak with a small margin of error at range from leak sensors.

[0005] Multiple factors (e.g. leak shape, pipe material, pipe thickness, operating pressure) determine the initial amplitude of the leak noise. As the sound of the leak travels to a leak sensor, the sound attenuates (loses volume) at a rate that is determined by factors such as pipe size, pipe material and soil type. For example, High Density Polyethylene (HDPE) has higher attenuation than that of a cast iron pipe. These factors play a complex role in determining the maximum range of passive sensing technologies.

[0006] Passive technologies, or data loggers are often used in combination with each other. A sensor can be installed every 100-300m along a network and continuously listen for the presence of a leak noise. Once a leak is heard by the sensor the utility may determine additional investigations are required. The utility may send out a crew to use a second technology to confirm the result, or reject it as a false positive, and they may also determine the exact location of the leak. This practice involves multiple difference sensors and a high degree of human intervention, meaning it is inefficient at a large scale.

[0007] With a large degree of human intervention, it is possible to increase the accuracy of passive sensors. One such example is to use correlating technologies. These technologies will determine the location of a leak. However, the user must place instruments on either side of the leak. This first requires the general location of the leak to be known. The type of pipe material must also be known and is usually manually entered into a correlating device. Another such example is a listening stick, which requires the operator to walk along the pipe network and listen for the leak noise. Again, this technology is passive and relies on listening to the noise created by the leak. It also requires a high level of intervention from an operator.

[0008] Passive technologies are also suspectable to false positives alarms (i.e. reporting a leak when none exists) as external events such as pumps, cars, or construction trigger noise events in pipes. Existing passive technologies also cannot determine a leak’s location around a corner in a complex network, for example an in bracket loop or where multiple in bracket paths exist.

[0009] A significant issue for remote monitoring is time synchronisation between sensors. Time synchronisation plays a significant factor in the level of accuracy in identification of a leak. Existing remote correlation systems use on-board clocks or quartz crystal clocks. Such clocks drift with respect to actual time meaningfully in time critical applications and the devices are required to re-synchronised to a master clock frequently, such as every few minutes, to maintain millisecond synchronisation accuracy. Even millisecond accuracy significantly limits leak location accuracy when at distances further away from the leak. For example, when the speed of sound is 1 ,200 m/s, a 1 millisecond difference can easily equate to a location difference of +- 1 .2 meters. [0010] Most correlators typically use local short range radio communication between a local base station and the sensors to synchronise the time between the sensors which adds additional costs and components to the system. Furthermore, such a technique is only suitable where the sensors are within communications range, which can be adversely affected by surrounding terrain and infrastructure, and even then, still inherently introduces communication delays that vary depending upon how far each sensor is located from the base station.

[0011 ] In the pursuit of high precision synchronisation to improve accuracy, the invention described in US patent no. 10,720,168 entitled ‘PPS Tagging of Acoustic Sample Data’ uses pulse-per-second (PPS) from a GPS receiver to synchronise recording devices by modifying upper bytes of recorded data to tag certain samples as PPS samples. Whilst such a technique provides significant advantages over previously described synchronisation methods, it has a few shortfalls when it comes to being able to obtain a theoretically possible high precision.

[0012] For example, the recorded data is converted to digital at each device. Samples taken at each device have two types, either a regular sample or a PPS sample (refer figure 4 of the US patent no. 10,720,168). These devices each have their own clock or oscillator which is prone to drift. Although marking ‘PPS samples' can be used to assist with sample alignment, it does not account for skews that may arise due to compression or decompression of the sample due to the analogue to digital (ADC) on the device running fast or slow.

[0013] Furthermore, each process performed by the microprocessor at the device (e.g. detecting PPS, processing and modifying samples, storing and retrieving sample data to RAM to modify, etc.) takes clock cycles which, when striving for submillisecond accuracy add meaningful delays between the PPS being identified and associated with a sample. This delay too will vary depending upon the drift of the processor’s clock. Such factors can introduce tangible timing inaccuracies limiting the accuracy in being able to pin point the location of a leak, for example. For the avoidance of doubt, the method described US patent no. 10,720,168 is not considered to be well known or common general knowledge, at least in the field of leak detection. OBJECT OF THE INVENTION

[0014] It is an object of the invention to overcome and/or alleviate one or more of the abovementioned problems and/or provide the consumer with a useful or commercial choice.

SUMMARY OF THE INVENTION

[0015] In one form, although not necessarily the only or broadest form, the invention resides in a method of synchronising sensor measurements, the method comprising: simultaneously recording one or more sensor measurements at a first sensor and one or more sensor measurements at a second sensor with a pulse per second (PPS) signal received at the processor of each of the sensors; storing simultaneously recorded sensor measurements with recorded PPS signal from the first sensor together and storing simultaneously recorded sensor measurements with recorded PPS signal from the second sensor together; and comparing the recorded PPS signals from the first sensor and the second sensor; and synchronise the sensor measurements the first sensor and the second sensor in accordance with an analysis of the comparison of the recorded PPS signals from the first sensor and the second sensor.

[0016] The method may comprise one or more further sensors. The method may comprise a plurality of sensors. One or more of the first sensor, the second sensor, and the further sensor(s) may be different to one or more of the others. One or more of the sensors may take acoustic vibration recordings. One or more of the sensors may comprise one or more acoustic vibration inputs of a hydrophone, an accelerometer, a geophone, microphone, a piezoelectric device, or the like.

[0017] Storing simultaneously recorded sensor measurements with recorded PPS signal from the first sensor together and storing simultaneously recorded sensor measurements with recorded PPS signal from the second sensor together may comprise each sensor storing its respective simultaneously recorded sensor measurements and PPS signal.

[0018] Stored simultaneously recorded sensor measurements with recorded PPS signal from each of the first sensor and the second sensor is communicated to an external device for processing. The external device may comprise a server. [0019] The steps of comparing the recorded PPS signals from the first sensor and the second sensor and synchronising the sensor measurements from each sensor in accordance with an analysis of the comparison of the recorded PPS signals from the first sensor and the second sensor may be performed at the external device.

[0020] The sensor measurements and the PPS signal may be recorded in separate audio channels or tracks of an audio recording. A recorded sensor measurement may be stored in a left channel or a right channel of a stereo audio recording and the PPS signal may be stored in the other of the left and the right channel of the stereo audio recording.

[0021 ] The PPS signal may be a satellite signal of 1 Hz or greater received by a satellite receiver at each sensor. The PPS signal may be a satellite signal from a global navigation satellite system (GNSS). The GNSS may be GPS or any other similar satellite system. The sensor may comprise a GPS receiver configured to receive the PPS satellite signal.

[0022] The received satellite signal may be transmitted over an analogue circuit in real time to an analogue to digital (ADC) input of, or in communication, a processor of the sensor. The ADC input may be of an integrated ADC audio codec package of, or in communication with, the processor. The analogue circuit may comprise a passive filter configured to condition a digital PPS signal into a suitable analogue waveform for digitising. The sensor measurements may be transmitted over an analogue circuit to another input of the ADC. The sensor measurements may be passively filtered in the analogue circuit. The sensor measurements may be amplified in the analogue circuit. The analogue PPS signal and one or more analogue sensor measurements signal may be simultaneously recorded as separate audio channels or tracks of a multichannel or multitrack audio recording by the ADC.

[0023] The PPS may be used to discipline a clock of the processor of each sensor. The PPS may be used to discipline the ADC audio codec package. The PPS may be used to calibrate all digital clocks and oscillators at each sensor.

[0024] In another form, there is provided a method of detecting a leak within a pipe network comprising: installing a plurality of sensors in the pipe network; receiving and recording one or more sensor measurements at each of the plurality of sensors in the pipe network; and synchronising sensor measurements from the plurality of sensors, preferably according to the hereinbefore described method.

[0025] The method may of detecting a leak within a pipe network may further comprise: emitting a signal through one or more of the plurality of sensors; receiving sensor measurements received by one or more of the other sensors; analysing the received sensor measurements to determine if there are any leaks; and determining the location of a leak based on the synchronised sensor measurements received through the sensors, and the locations of the sensors.

[0026] Determining the leak location may also be based on a calculated time delay of a leak signal from one or more of the sensors. Each sensor may be activated according to one or more of a schedule and detection of a transient event occurring.

[0027] The method of detecting a leak within a pipe network may further comprise: positioning sensors by: i) dividing an area of a pipe network into discrete segments; ii) uniquely identifying each pipe segment with a pipe segment identifier; iii) increasing the number of sensors in the pipe network by one; iv) determining the number of locations the sensors can be installed in the pipe network; v) positioning the sensors throughout the pipe network between selected discrete segments; vi) for each pipe segment, determining the distance from each pipe segment to each sensor and utilising the results from each sensor to form a sensor-distance identifier; vii) checking that no pipe segment has the same sensordistance identifier; viii) if any sensor-distance identifier for any pipe segment is not unique, then relocate the sensors throughout the pipe network between selected discrete segments and going back to step (v); and ix) if any pipe segments are not being monitored by a sensor then increasing the number of sensors in the pipe network by one and going back to step (iv) until the all pipe segments are monitored.

[0028] In another form, there is provided a method of detecting a leak within a pipe network including the steps of: identifying sensor installation locations; installing a plurality of sensors; calibrating one or more of the plurality of sensors; emitting a signal into the water or pipe wall; identifying and tracking naturally occurring transient pressure waves in the water network; receiving signal data and background data by one or more of the other sensors; determining a leak is present by using a trained machine learning algorithm; and determining the location of the leak based on the signal data received through the sensors, the location of the sensors, and the time synchronisation between the sensors.

[0029] The step of receiving signal data may comprise recording the signal data. The step of determining the location of the leak may comprise comparing received signal data and timing from a plurality of sensors. The step of comparing received signal data from a plurality of sensors may comprise identifying leak locations that most align. The step of determining the location of the leak on the utilities network may comprise using triangulation and/or pipe route pathing (Pipe route pathing is the process of determining which pipe path the leak exists on between ‘In bracket’ sensors, including the discriminating a specific pipe in the case where multiple parallel pipe paths exist between the sensors).

[0030] The location of the sensors may be determined by several methods. For example, a sensor may be located on a discrete length of pipe. Another example is to locate a sensor based on a grid which divides a pipe network into segments and locating a sensor within each segment. Preferably the step of identifying sensor installation locations includes optimising the number of sensors required to provide full coverage of the pipe network. Optimising the number of sensors required to provide full coverage of the pipe network may comprise using an optimisation strategy that identifies quasi-optimum locations.

[0031 ] Typically each sensor is calibrated using a signal frequency. Each sensor may be calibrated by conducting a signal frequency sweep each to identify the optimal transmission signal frequencies from the sensor being calibrated to one or more adjacent sensors.

[0032] The sensor sweep frequencies may extend from 10Hz to 1000Hz. Preferably, the sensor sweep frequencies extend from 20Hz to 500Hz. Even more preferably, the sensor sweep frequencies extend from around 30Hz to around 400Hz.

[0033] Typically each sensor is calibrated using a signal amplitude. Each sensor may be calibrated by conducting a signal amplitude sweep to identify optimal transmission signal amplitudes from the sensor being calibrated to one or more adjacent sensors. The signal amplitude is preferably constant. The signal amplitude is typically attenuated at different rates depending on the pipe it is traversing. The frequency with the highest amplitude received at the adjacent sensor correspond to a preferred calibration signal frequency.

[0034] The emitted signal may be generated through any device such as a rotating eccentric weight attached to the pipe, a speaker mounted on the pipe, a valve connected to the pipe, a tap on the pipe, or other conceivable methods. In a preferred form, the emitted signal may be generated using a device comprising a plunger or diaphragm operatively connected to the pipe network.

[0035] The emitted signal may comprise a reference signal in the form of a chirp. The chirp may be a quadratic chirp. The chirp may range from 10Hz to 100Hz, preferably from 20Hz to 80Hz, more preferably from 40Hz to 60Hz, and even more preferably from around 50Hz. The chirp may range to 100Hz to 5kHz, preferably to 200Hz to 1 kHz, more preferably to 400Hz to 600Hz, and even more preferably to around 500Hz. The chirp may span a time period of between 1 second and 10 seconds, preferably a time period of between 2 and 8 seconds, more preferably a time period of between 3 and 7 seconds, and even more preferably a time period of around 5 seconds.

[0036] The received signal may be pre-processed to remove unwanted noise, and/or amplify desired frequencies to increase leak detection range. Digital signal processing may be used for this purpose.

[0037] An algorithm may be used to analyse the data and/or to identify the presence of one or more leaks. The algorithm may comprise a machine learning algorithm. The machine learning algorithm may comprise a time frequency neural network (TFNN). Pre-existing anomalies, such as pipe topography, may be identified during calibration. Newly identified anomalies, such as from a leak, may be identified post-calibration. Anomalies may also be able to be identified during calibration.

[0038] The measured speed of sound in water may be used to calculate the distance from one or more sensors to any anomalies or leaks.

[0039] Each possible leak location may be determined based on the time delay of a leak signal from one or more sensors.

[0040] For each leak signal, the leak location may be determined by triangulation and/or pipe route pathing. A plurality of possible leak locations may be identified for each sensor. The leak location may be identified by determining leak locations that align from two or more of the sensors. A leak location determined to be on a sensor may represent an out of bracket leak (e.g. unable to path specific leak position beyond the location of the sensor).

[0041 ] The method may comprise the step of recalibration. Recalibration may occur periodically, and/or after an anomaly has changed such as a leak having been repaired or pipe network topography being altered. Recalibration preferably comprises substantially the same steps as calibration, but performed to record speed of sound changes.

[0042] The method may further comprise the step of synchronising the plurality of sensors. The sensors may be synchronised to less than 1 millisecond (ms) of each other, preferably less than 50 nanoseconds (ns) of each other, and even more preferably to within 20 ns of each other. Under ideal circumstances the sensors may be synchronised to within around 1 ns of each other. The step of synchronising the plurality of sensors may comprise receiving a satellite signal. The satellite signal may comprise receiving a GPS signal. It should be appreciated that other suitable Global Navigation Satellite System (GNSS) signals may be used instead or in addition to GPS and no limitation is meant thereby.

[0043] The method may further comprise the step of recording a satellite signal. The satellite signal may comprise a pulse per second (PPS) signal. The step of recording a satellite signal may be performed simultaneously with the step of receiving signal data received by one or more of the other sensors. The step of analysing the received signal data to determine if there are any leaks may further comprise analysing the received signal data in conjunction with the simultaneously recorded satellite signal. [0044] The sensor according to the method may comprise a GPS disciplined clock onboard that is configured to allow remote time synchronisation of the sensors independent of their proximity to each other.

[0045] In another form, the invention may reside in a method of positioning sensors within a leak detection system including the steps of: i) dividing an area of a pipe network into discrete segments; ii) uniquely identifying each pipe segment with a pipe segment identifier; iii) increasing the number of sensors in the pipe network by one; iv) determining the number of locations the sensors can be installed in the pipe network; v) positioning the sensors throughout the pipe network between selected discrete segments; vi) for each pipe segment, determining the distance from each pipe segment to each sensor and utilising the results from each sensor to form a sensordistance identifier; vii) checking that no pipe segment has the same sensor-distance identifier; viii) if any sensor-distance identifier for any pipe segment is not unique, then relocate the sensors throughout the pipe network between selected discrete segments and going back to step (v); and ix) if any pipe segments are not being monitored by a sensor then increasing the number of sensors in the pipe network by one and going back to step (iv) until the all pipe segments are monitored.

[0046] The starting number of sensors is typically one, for leak determination over a given zone of network piping, however a minimum of two sensors is required to accurately calculate leak position within this zone. However, for large pipe networks the starting value may be larger than one such as two, three, four, or five or more.

[0047] The discrete pipe segments may be of different lengths. The pipe segments are typically between 1 and 20 metres in length. It should be appreciated, however, that pipe segments could be of shorter or longer lengths such as, for example, between 0.5 and 50 metres.

[0048] The sensors may be the same as each other. Preferably, each sensor can sense a plurality of segments. Each sensor may be able to detect over 100m, preferably over 500m, and even more preferably over 1 km, depending on the sensing input instrument. [0049] The step of utilising the results from each sensor to form a sensor-distance identifier may comprise combining the results from each sensor to form the sensordistance identifier. The step of combining the results from each sensor may comprise concatenating each sensor distance to form the sensor-distance identifier.

[0050] Further features of the invention will become apparent from the detailed description below.

BRIEF DESCRIPTION OF THE DRAWINGS

[0051 ] Embodiments of the invention will be described, by way of example only, with reference to the accompanying figures in which:

[0052] Figure 1 illustrates a junction of a pipe network with a single sensor identifying multiple leak solutions;

[0053] Figures 2A to 2D illustrate an example pipe network split into sections with possible sensor deployments;

[0054] Figure 3 illustrates an example pipe network with multiple sensors and leaks; [0055] Figure 4 illustrates a measured frequency response of a 250mm LIVPVC pipe; [0056] Figure 5 illustrates an example installation of different pipe types with sensors placed at the ends and midpoint;

[0057] Figure 6 illustrates an emitted reference signal and a received return signal;

[0058] Figure 7 illustrates a demodulation of the signals of figure 6;

[0059] Figure 8 illustrates an example output form illustrating identification of leaks;

[0060] Figure 9 illustrates an example of how multiple possible anomaly locations on a pipe network can exist with a single sensor;

[0061] Figure 10 illustrates an example of how multiple sensors can be used to determine a single anomaly location for the pipe network of figure 5;

[0062] Figure 11 illustrates an example of multiple sensors being utilised to find an anomaly in a pipe network comprising different pipe materials;

[0063] Figure 12 illustrates a representative diagram demonstrating clock speed skew for an analogue to digital converter (ADC);

[0064] Figure 13 illustrates a two channel (stereo) audio recording with an analogue filtered PPS signal stored in one channel and a sample sinusoidal sensor measurement store in the other channel; and

[0065] Figure 14 is a block diagram illustrating a sensor synchronisation process. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0066] A method of creating a leak detection system that can both be installed permanently across a pipe network to proactively look for leaks, or to be used as a reactive leak detection device, is provided. To create a comprehensive leak detection system over a pipe network, it is important that all pipes located within the pipe network can be monitored for leaks. Positioning of sensors within a pipe network that optimises both the number of sensors required to monitor the network and the position of each of the sensors such that the position of any leaks can also be determined is also provided.

[0067] Figure 1 illustrates a single sensor 100 installed near a junction 120 of a pipe network. This single sensor 100 cannot differentiate signals returning from the left path 122 or the right path 124. This results in multiple solutions when a single sensor identifies a leak. In this simple example, the possible solutions from sensor 100 could be a leak at 126, a leak at 128, or leaks at both 126 and 128. One way to resolve this issue is to install another sensor at a different location and compare their signals.

[0068] In the simple situation illustrated in figure 1 , a further sensor located on either left path 122 or right path 124 would be sufficient. However, in more complex real world scenarios many sensors may be required to ensure unique solutions can be provided at all locations. As there is a cost associated with installing sensors (both in terms of hardware and installation) it is desirable to try to optimise the sensor install locations to the fewest sensors that will yield unique solutions across the pipe network. [0069] Figure 2A shows an example pipe network with several different branches. A first step in locating sensors within such a pipe network is to divide the pipe network into segments as shown in figure 2B. The length of each segment is less that the sensing distance of a sensor. In this example, each sensor has a range of five segments of pipe. In normal use, sensors can typically be located within several defined locations such as inspection openings. However, for the purpose of this example, it is assumed that the segments are of equal length and the sensors can be located between each and every segment. In reality, this will likely not be the case. However, the methodology is still applicable. Once the segments have been determined, each segment is allocated a segment identifier in the form of a segment number also shown in figure 2B (segment nos.1 to 11 ).

[0070] It would likely be clear to a person skilled in the art that two sensors would not be sufficient to cover such a pipe network. It may not be so clear if three sensors would be sufficient. Accordingly, for this example, the starting number of sensors is three, which each have unique sensor identifiers in the form of sensor numbers. The sensors are located within pipe network at spaced location which appear to cover the network as shown in figure 2C.

[0071 ] The distance of each pipe segment from each sensor is then measured. Table 1 (below) shows the distance of each pipe segment from each sensor for the position of the sensors shown in figure 2C. The distance from each sensor is then used to create a concatenated sensor-distance identifier shown in the last row of the table. This example assumes each sensor can monitor up to four lengths of pipe. A “0” result is recorded if a pipe segment is further away (i.e. 5 segments or greater). A result of “000” would indicate that the pipe segment is not being monitored (i.e. out of range of all sensors). In this example, all of the pipe segments are able to be monitored by three sensors. If one or more of the pipe segments returned a “000” result then either rearrangement of the sensors or an additional sensor would be required.

Table 1

[0072] In table 1 , the sensor-distance identifier is the same for pipe segment 6 and pipe segment 10 (i.e. 332). Accordingly, if a leak were to occur in pipe segment 10, it would not be clear from the information obtained from the sensors if the leak had occurred in pipe segment 6 or pipe segment 10. Accordingly, the locations of the sensors should to be optimised as shown in figure 2D for example. Table 2 (below) shows the distance of each pipe segment from each sensor for the position of the sensors shown in figure 2D.

Table 2

[0073] The change in sensor positions changes the sensor-distance identifiers for each of the pipe segments as shown in Table 2. Each of the pipe segments now has a unique sensor-distance identifier. Accordingly, the location of any leak that occurs in any segment can be identified without having multiple segment possibilities. Once optimum sensor locations have been determined, the sensors are then installed within the pipe network.

[0074] For a large and complex pipe network, performing the aforementioned sensor optimisation may be computationally onerous. There are numerous issues in attempting to efficiently optimise sensor locations such as: initial placement of the first sensor may lead to a suboptimal solution, placement of next sensors may affect the effectiveness of previous sensors, junctions require then number of legs (1 sensors for unique paths), larger networks will require more sensors, total number of sensors are unknow, can only know by testing all possible combinations, a given solution can be evaluated in polynomial time and compared to previous solutions. It is a Nondeterministic Polynomial - Complete problem (NP- Complete problem) meaning that for which no efficient solution algorithm has been found. A sensor position placement solution may therefore be derived using an optimisation strategy that identifies quasi-optimum sensor locations with a reduced computational burden.

[0075] Figure 3 illustrates an example pipe network with sensors Sn, leaks Ln, and junctions J _n . A path is from leak to junction, junction to junction and junction to sensor or leak to junction where there is no junction to between leak and sensor.

[0076] A unique path is where the path to the sensor(s) from a leak follows different segments in the network, for example from leak three (L3) to sensor two (S2) the path would be L3 -> S3 not passing through any junctions and the path to sensor one (S1 ) would be L3 -> J3 -> S1 ; as it passes through junction 3 (J3). With the distance calculated from these two sensors we can calculate the location of leaks between J3 and S2, J3 and J2, J3 and J4 and we have a unique path for this pipe section.

[0077] For the pipe network illustrated in figure 3, and assuming all sensors have sufficient range, sensor to leak paths are as follows:

51 -> J2 -> J1 -> L1 , D=7

S1 -> J2 -> L2, D=4

S1 -> J3 -> L3, D=7

S1 -> J3 -> J4 -> L4, D=7

51 -> J3 -> J4 -> J5 -> J6 -> L5, D=14

52 -> J3 -> J2 -> J1 -> L1 , D=16

52 -> J3 -> J2 -> L2, D=12

52 -> L3, D=2

S2 -> J3 -> J4 -> L4, D=6

52 -> J3 -> J4 -> J5 -> J6 -> L5, D=13

53 -> J4 -> J3 -> J2 -> J1 -> L1 , D=16

53 -> J4 -> J3 -> J2 -> L2, D=13

S3 -> J4 -> J3 -> L3, D=6

53 -> L4, D=2

53 -> J4 -> J5 -> J6 -> L5, D=11

54 -> J5 -> J4 -> J3 -> J2 -> J1 -> L1 , D=17

54 -> J5 -> J4 -> J3 -> J2 -> L2, D=14

S4 -> J5 -> J4 -> J3 -> L3, D=7

S4 -> J5 -> J4 -> L4, D=5

S4 -> J5 -> J6 -> L5, D=8

[0078] Leaks L3 and L4 are the only leaks with unique paths, with paths to/from S1 , S2, S3 and S4, but S3 and S4 travel the same path [J4 -> J3 -> L3] to L3 and can only be counted as 1 part of the unique solution. S3 and S4 cannot discriminant if leak three (L3) is two measurement units (D) to the West or South of junction 3 (J3).

[0079] The optimisation strategy assesses the network with test points (simulated leaks) that include: junctions, dead ends, even distances from junctions (e.g. 1 m), centres of segments between junctions. Further test points can be added such that the distance between test points is greater than the maximum of the sensor range. Simulated sensors are placed at 1/5 of the maximum sensor range along the pipe network.

[0080] The path and distance from every test point to every sensor location is determined and sensor location to sensor location distance (via pipe length) is determined. Possible paths between test points and sensor locations are reduced to unique paths only. Sensor locations are ranked according to a number of unique paths. The sensor location with the most unique paths is designated as a first sensor. [0081 ] Sensors in range of the first sensor are then identified. Each of the in-range sensor locations are tested iteratively with respect to the previous sensor for the number of test points that have:

A. 0 sensors in range

B. 1 sensor in range with unique distances

C. 1 sensor in range with duplicate distances

D. 2 sensor in range with unique distances to each

E. 2 sensor in range with duplicate distances to each

F. 3 sensor in range with unique distances to each

G. 3 sensor in range with a mix of unique and duplicate distances to each

H. 3 sensor in range with duplicate distances to each

I. 4+ sensor in range with unique distances to each

J. 4+ sensor in range with a mix of unique and duplicate distances to each

K. 4+ sensor in range with duplicate distances to each

[0082] The sensor location that maximises the distance and minimises the rest is selected and set as the location for the next sensor. Sensors in range of the next sensor are then identified. Each of the in-range sensor locations are tested iteratively with respect to the previous sensor for the number of test points that have A to G listed above. This process is repeated until A is zero at which point a base placement with a maximum number of sensors for the selected area is provided.

[0083] The base placement may then be optimised. An attraction force (AF) for each sensor is determined by summing the distance to each test point within range and a repulsion force (RF) for each sensor is determined by summing the distance to each other sensor within range. A sum force (SF) is determined by summing the AF and RF. If the SF is greater than or equal to the distance of a neighbouring sensor placement, the location of the present sensor is moved to that location and this process is repeated until a steady state is found. The resultant sensor locations are optimised (or quasi-optimised compared to a ‘perfect’ solution to the NP-Complete problem).

Frequency Sweep

[0084] Once the sensors have been installed, or as they are installed each of the sensors should be calibrated. Sound waves are attenuated at different rates as they travel through water pipes based on properties such as pipe wall thickness, density, and size. Based on the applicant’s experiments, it emerged that preferred frequencies appear between 30Hz - 400Hz, and typically have a bandwidth of between around 6Hz - 10 Hz. Signals outside of these preferred frequencies were attenuated.

[0085] Figure 4 illustrates how frequencies ranging from 20Hz - 500Hz were attenuated over a specific 250mm Unplasticized Polyvinyl Chloride (UPVC) pipe. In this example, an attenuation of approximately -60dB per decade was observed. For comparison, a 200mm Asbestos Cement (AC) pipe was found to have a preferred frequency range of 52Hz ± 8Hz and similar attenuation rates as the UPVC pipe.

[0086] The narrow bandwidth presents a challenge when a pipe network includes multiple pipe types. An example of this issue is shown in Figure 5, where a 250mm UPVC pipe 10 is connected to a 200mm AC pipe 20. Sensors 1 , 2, and 3 are located at the start (sensor 1 ), midpoint (sensor 2), and end (sensor 3) of a length of pipe having the two different types of materials. Based on the length of each pipe type, the preferred frequencies, and attenuation rate, it is possible that no signal from sensor 1 will reach sensor 3.

[0087] After being installed each sensor (1 -3) conducts a frequency sweep between, for example, 20Hz and 500hz. The results from nearby sensors are recorded as shown in Table 3, below, for each combination of the sensors shown in figure 5. It should be appreciated that each sensor may both transmit and receive sound.

Table 3 [0088] When location 2 injects the signal it completes the step twice. The first being at the preferred frequency of around 36Hz for the LIPVC portion of the length of pipe and the second being at the preferred frequency of 52Hz for the AC portion of the length of pipe. Ideally, any sensor that is located between different pipe types should emit multiple frequencies to account for the preferred frequencies of each pipe type. Sensors 1 and 3, on the other hand, only need to complete the step once as they are located at portion of pipe that is of the same type. This process is desirable to ensure that the pipe network receives comprehensive coverage even when the network comprises different pipe materials.

[0089] Where the leak detection use case is correlation based between a pair of sensors, and the distance between the two sensors is known, and the signal has been transmitted and received, the time-based measurement can be used to calculate the speed of sound in the fluid between the two sensors for that pipe segment. Standard correlation may then be used to determine the time taken to transmit and receive the signal between sensors, with the correlated time difference being referenced from the emitting sensor. This would lead to a calculated speed of sound being:

Measured Speed of Sound = Distance between sensors/correlated time difference.

[0090] This measured speed of sound can then be applied to correlate and determine the leak location between the sensors with more accuracy than using an estimated speed of sound that is based on pipe material profile.

[0091 ] Alternatively, where the use case is a singular sensor to detect leaks, once a signal has been transmitted and received, it can be converted from a time-based measurement to a distance-based measurement by using the speed of sound in the fluid, and considering the travel time of the injected signal. For example, assuming for this example that the pipe network is carrying water, that an estimated speed of sound in water is 1500m/s, and that the injected signal must travel to the leak site and a response must make the return trip, then a result recorded at 0.19 seconds, as illustrated in figure 8, would translate to a distance of 142.5 metres as follows:

0.19 seconds x 1500 metres per second

- - - = 142.5 metres

2

A result recorded at 0.43 seconds would be:

0.43 seconds x 1500 metres per second

- - - = 322.5 metres

2 This measured speed of sound can then be applied to correlate and determine the leak location between the sensors with more accuracy than using an estimated speed of sound that is based on pipe material profile.

[0092] For the use case of a single sensor detecting leaks, it has been found that there are advantages in injecting a more complex reference signal. The reference signal may be chosen based upon known properties of the pipe network including at least one of pipe material, pipe diameter, age, and substrate in which the pipe is buried. It has been found that a reference signal in the form of a ‘chirp’ is usually suitable. Figure

6 illustrates a reference signal in the form of a quadratic chirp 50 ranging from 50Hz to 500Hz over five seconds and a response signal 52 received two seconds after the reference signal emission started.

[0093] These signals can be processed to extract information about the pipe network. The reference signal can be treated as a carrier wave, and one or more of time delay, frequency response, and phase shift of the response signal can be determined. Figure

7 illustrates demodulated signals comprising a summed signal 54 (which is a quadratic) and an absolute value difference signal 56 (which is ‘V’ shaped). The shape (angle) of the difference signal 54 varies based on distance. Advantageously, it is a shape that does not arise from other external phenomenon (e.g. naturally or from electricity transmission) so is readily identifiable.

[0094] Multiple demodulated and combined signals can be statistically analysed to reduce noise, lower dimensionality, and eliminate conflating variables. This produces a sample in the feature space know as a feature. The combined signals, demodulated signals, and features can be input into one or more machine learning models. The machine learning models may be an ensemble of supervised and unsupervised machine leaning models. With reference to tagged and untagged data, and a database of information about pipe networks, a distribution of predictions about the physical properties of the pipe network can be made.

[0095] The prediction distribution can be fed into a geographic information system (GIS) and used to highlight regions of interest such as leaks, closed valves, material or geometry changes, blockages or regions of pipe wall thinning.

Triangulation

[0096] Typically, there are multiple possible solutions when a sensor that is operating in isolation from any other device identifies the presence of a leak. While the sensor may report a leak distance (e.g. 200m away), the complexity of the pipe network may result in multiple possible leak locations. Such a situation is illustrated in Figure 9 whereby three possible leak locations exist a measured distance from the sensor.

[0097] A simplified version of an algorithm that may be used to identify a unique solution is to:

(i) assign a location and unique reference identification to each possible leak location;

(ii) for each possible leak location identified by each sensor:

(a) calculate a midpoint and each other possible location from the other sensors; and

(b) calculate a distance from the midpoint to itself;

(iii) whereby the midpoint with the smallest distance is indicative of the actual leak location.

[0098] Measurements from multiple sensors may be utilised to narrow down the possible leak locations to a single unique solution as illustrated in figure 10. As can be seen, each sensor determines a plurality of possible leak locations with the actual leak being indicated by an aligned solution from both sensors (i.e. possible location no. 1 ).

[0099] A more complex version of this algorithm may incorporate how different signals attenuate at different rates depending on factors such as pipe material. Figure 11 illustrates how this additional information may result in a different outcome.

[0100] Table 4, below, shows example results from applying the algorithm to the networks illustrated in figures 9 and 11. In this example a leak should be identified as the midpoint between sensor 1 location 1 and sensor 2 location 2 for figure 10 and sensor 1 location 2 and sensor 2 location 2 for figure 11 as these midpoint distances are the smallest.

Table 4 Synchronisation

[0101 ] In order to minimise margin of correlation error, and to accurately measure the speed of sound, and accurately determine the location of a leak, a sensor synchronisation with sub-millisecond second accuracy has been found to be highly desirable. To achieve this, the system described may use a Global Navigation Satellite System (GNSS) such as GPS(or other similar satellite constellation) disciplined clock that stays in synchronisation without having to be physically connected. All digital clocks and oscillators at a sensor (e.g. a processor, multiprocessing unit (MPU), ADC codec package, etc.) can be disciplined/synchronised periodically using a suitable satellite PPS signal. However, this does not account for any sensor clock drift between synchronisations.

[0102] A satellite (e.g. GPS) pulse per second (PPS) signal can be added to a sensor sample recording as a synchronisation track or channel. The synchronisation track or channel allows for alignment of leak data measurement samples to be within submillisecond (nano-seconds) of each other across a plurality of sensors prior to correlation, allowing for greater precision and accuracy. The sampling rate effectively becomes the limiting factor for accuracy.

[0103] Figure 12 illustrates erroneous analysis that can arise due to an inaccurately disciplined processor/ADC clock through an example of digitising a 5Hz sine wave. The first waveform shows an expected sample rate of 60 samples/sec. The second waveform shows a sampling rate where an analogue to digital converter (ADC) is running slightly faster than expected, e.g. at 62 samples/sec. The third waveform illustrates a resultant erroneous digitised signal which, when played back at 60 samples/sec, provides an incorrect wave frequency of 4.84Hz rather than 5Hz. Accordingly, even if the starting point of measurement is synchronised, an inaccurate clock rate at a processor/ADC can result in skewing of the sample. For the avoidance of doubt, this is a simplified example and the 5Hz sine wave and selected sampling rates are not considered to be indicative of actual signals and preferred sampling rates in use.

[0104] According to a form, a high precision pulse per second (PPS) satellite signal is recorded as a synchronisation track to enable a time synchronisation of signal recordings across multiple disconnected devices. The sensors may comprise a GPS disciplined clock that enables remote time syncing of devices independent of their proximity to each other. [0105] Figure 13 illustrates a two channel (stereo) audio recording with an analogue filtered PPS signal stored in one channel (left channel) and a sample sinusoidal sensor measurement stored in the other channel (right channel). The input PPS signal’s square waveform illustrated in the left channel when sampled may have a minor amount of distortion (over/ undershoot) at the comers of the pulse. The amount of distortion that occurs will vary depending upon the sampling rate used. A sufficiently high sample rate is selected to minimise this distortion. Additionally the distortion does not significantly impact the timing synchronisation, as in practice the instantaneous rising edge of the recorded PPS signal is aligned with the PPS input pulse rising edge, with a referential maximum lag of one sample period. Again, this lag is minimised by selecting a sufficiently high sample rate. From such a two channel recording the sensor measurement can always be corrected for clock errors, such as that illustrated by figure 12, at later date by analysing the analogue recorded PPS signal. Of most importance is the rising edge of the PPS signal, but it should be appreciated that other characteristics of the analogue recorded PPS signal may be utilised to determine any errors in the sampling rate (e.g. due to clock drift).

[0106] Figure 14 is a block diagram illustrating a sensor synchronisation process at a sensor. A sensor measurement is recorded 100, conditioned 120, and inputted into a first (e.g. right) channel of a two channel (stereo) ADC/Codec package 250. A satellite 200, such as GPS, provides a satellite signal such as PPS to a processor clock 210 of a processor 220 and to the ADC/Codec clock 230 of an ADC/Codec package 230. The clocks 210, 230 are disciplined (synchronised) from the satellite signal. For the reasons outlined above, simple synchronisation of on device clocks cannot completely eliminate clock drift and resultant sampling skew. The PPS signal from the satellite 200 is therefore conditioned (filtered) 240 into an analogue signal and inputted into the other channel (e.g. left) of the ADC/Codec package 250. Both the conditioned analogue sensor measurement 120 and conditioned PPS signal 240 are input into the ADC/Codec package 250 simultaneously and recorded/saved 260 together in a single two channel audio file. As both channels utilise the same ADC/Codec clock, should any drift have occurred this would be identifiable as a skew in the conditioned PPS signal 240 which can then be analysed to correct any skew in the sensor measurement recording. Advantages

[0107] Advantageously, the present invention provides a method of synchronising sensor measurements to a very high accuracy, preferably under 5ns. This increased synchronisation precision allows for greatly improved positional determination in a pipe network leak detection system. However, it should be appreciated that the method of synchronisation can have other applications. It is particularly well suited to synchronising analogue sensor measurements, such as those described herein, but again no limitation is necessarily meant thereby as digital signals can also be conditioned to be suitable for analogue capture and subsequent digital recreation (if necessary).

[0108] The plurality of sensors can receive passive leak noises and the location can be determined very accurately using the same time delay functionality with synchronised sensor measurements.

[0109] In a preferred application, the present invention can provide a method to identify and accurately pinpoint a fluid leak location in a pipe network. The invention may use predetermined discreet frequencies and amplitudes of an emitted sound wave depending on the characteristics of the network. The signal travels down the pipe, and when it encounters a leak, a bend in the pipe, a T-junction or some other anomaly in the pipe, a portion of the sound wave is reflected and refracted from these anomalies. These changes can be identified and a location determined using a time delay due to the speed of sound on the signals received by a plurality of sensors.

[0110] Furthermore, narrowband post processing techniques, such as lock-in amplification, to identify a correct signal can be implemented. These techniques can operate even if the signal to noise ratio of the detected signal is less than 1 . As a known sound wave can be emitted at a known time, the invention may be largely immune to extraneous events occurring outside the time or frequency band. Additionally, the shape of an emitted signal is typically known; therefore, the invention can exclude results that do not match an expected shape. These factors combine to prevent events such as cars, or construction, causing false positive results.

[0111 ] Further, false positives can be prevented through taking a passive sample over a period of time to ensure continuity of leak noise and/or using an Al algorithm to classify the noise.

[0112] By having an emitter or sending a signal into or through a pipe, the invention can automatically determine the speed of sound in actual conditions, thereby not requiring the properties of each pipe to be entered manually to determine the distance of the leak from the sensor, however theoretical speed of sound information may also be used.

[0113] Although the sensor synchronisation invention has primarily been described with respect to leak detection and, in particular, determining the location of a leak, it should be appreciated that it may be utilised in relation to other fields that require high precision synchronisation of measurement signals made by a network of independent sensors.

[0114] In this specification, the terms "comprise", "comprises", "comprising" or similar terms are intended to mean a non-exclusive inclusion, such that a system, method or apparatus that comprises a list of elements does not include those elements solely but may well include other elements not listed.

[0115] In this specification, terms such as upward, downward, horizontal and vertical, and their grammatical derivatives, are used to describe the invention in its normal orientation and are not to be construed to limit the invention to any particular orientation.

[0116] The reference to any prior art in this specification is not, and should not be taken as, an acknowledgement or any form of suggestion that the prior art forms part of the common general knowledge.

[0117] It should be appreciated that various other changes and modifications may be made to the embodiments described without departing from the spirit or scope of the invention.

[0118] In this specification, the terms "comprise", "comprises", "comprising" or similar terms are intended to mean a non-exclusive inclusion, such that a system, method or apparatus that comprises a list of elements does not include those elements solely but may well include other elements not listed.