Water leakage detection system

Abstract

The invention relates to a method for detecting leakage in a domestic water channel, comprising: (a) acquiring and recording a pressure signal S from the water channel; (b) defining a check period T and a sampling rate s, and sampling the signal S during the period T, in the sampling rate s; (c) defining a sliding window W having a duration t; (d) activating sliding of window W to slide within period T and to acquire, in each position of W, a collection of N time-domain vectors Vt from signal S; (e) for each possible pair m,n within the collected vectors Vt, calculating a respective distance value D m,n ; (f) given the plurality of values D m,n , assigning each value D m,n to a respective pixel within a synthetic image I; and (g) feeding the synthetic image I into a flow versus no-flow classifier to determine whether the synthetic image I is an I nf synthetic image indicating a no flow state, or an I f synthetic image indicating a flow state of the water channel.

Description

[0001] Water Leakage Detection System

[0002] Field of the Invention

[0003] The field of the invention generally relates to water leakage detection systems, particularly those designed to detect water leaks in domestic properties.

[0004] Background of the Invention

[0005] Fresh water is a precious resource that has experienced increased scarcity in recent years due to global warming, increased population, and pollution. Given this scarcity, water suppliers have increased their efforts to minimize leakage and water losses; however, this is generally done up to the water meter of each property, i.e., the range within the water supplier’s responsibility. Beyond this range, each property owner is responsible for water losses, if they exist, and pays for these water losses due to leakage.

[0006] Recent research estimates that an average home in the US loses almost 10,000 gallons of water yearly, and 10% of the homes suffer from a leak of 90 gallons or more daily. The average homeowner generally is unaware of leaks unless they are seen or amount to a level that damages the property’s structure. Most consumers are not provided with any means for checking and measuring water leakage beyond the monthly water bills. However, those monthly bills are often overlooked unless they include a considerable deviation from the regular water consumption measured by the supplier’s main valve’s flow meter.

[0007] Some prior art techniques have suggested the detection of water leaks based on microphones pressed against a selected water pipe and determining leakage if continuous water flow occurs at times expected for no consumption. However, this technique is prone to ambient noise and is unreliable.

[0008] US 11,493,371 suggests (a) using one or more pressure sensors outside of a direct path to continuously acquire a pressure signal and (b) a template-based classifier for determining from said acquired pressure signal a specific water device consuming water at a given time, based on thresholds similarity and a plurality of complementary distance metrics used to filter potential templates.

[0009] US 11,709,108 discloses a system that is configured to [i] measure the water pressure in cold and hot water lines proximate to a fixture of a water system of a structure, and (ii) generate pressure measurement data representing the pressure of the water. The system also includes one or more processors configured to detect a non-cyclical pressure event corresponding to a water leak in the water system of the structure during a first time period based on an analysis of information comprising the pressure measurement data. Both US 11,493,371 and US 11,709,108 suffer from the drawbacks of relatively low accuracy of leakage detection; their techniques are specific to a type of leak or cyclical versus non-cyclical pressure events or, for example, a particular frequency range that represents a specific fixture. These are too specific and cannot cover all fixtures and leaks without specific adaptations.

[0010] One object of the present invention is to provide an accurate, automatic, and reliable system for continuously monitoring and detecting water leakage in a property.

[0011] Another object of the invention is to provide a leakage detection system that is installable almost anywhere within the property’s water network.

[0012] Another object of the invention is to provide a leakage detection system that does not require any calibration during installation.

[0013] Another object of the invention is to provide a fully automated leakage detection system capable of detecting a leakage utilizing a pre-manufacturing technique.

[0014] Other objects and advantages of the invention will become apparent as the description proceeds. Summary of the Invention

[0015] The invention relates to a method for detecting leakage in a domestic water channel, comprising: (a) acquiring and recording a pressure signal S from the water channel; (b) defining a check period T and a sampling rate s, and sampling the signal S during the period T, in the sampling rate s; (c) defining a sliding window W having a duration t; (d) activating sliding of window W to slide within period T and to acquire, in each position of W, a collection of N time-domain vectors Vt from signal S; (e) for each possible pair m,n within the collected vectors Vt, calculating a respective distance value Dm,n; (f) given the plurality of values Dm,n, assigning each value Dm,n to a respective pixel within a synthetic image I; and (g) feeding the synthetic image I into a flow versus no-flow classifier to determine whether the synthetic image I is an Inf synthetic image indicating a no flow state, or an If synthetic image indicating a flow state of the water channel.

[0016] In an embodiment of the invention, each of the N time-domain vectors Vt is converted to a frequency domain vector Vf, and wherein the calculations of the respective D vectors are calculated between each possible pair m,n of the frequency-domain vectors Vf.

[0017] In an embodiment of the invention, the method further comprising repeating steps (a) and (d) -(g), thereby to repeatedly determine no-flow state synthetic images Inf or flow state synthetic images If within the channel.

[0018] In an embodiment of the invention, the flow versus no-flow classifier is a machinelearning, neural-based classifier.

[0019] In an embodiment of the invention, the method further comprising counting sequential no-flow synthetic images, and alerting for a leakage if the counted number of sequential no-flow-synthetic images does not reach within a predefined global period a threshold number 7? at least a predefined number of times Q. In an embodiment of the invention, the given global period is in the order of at least 4 hours, and wherein the number of occurrences Q depends on said given global period.

[0020] In an embodiment of the invention, the given global period and the threshold number R, and the number of occurrences Q are tunable by a user.

[0021] In an embodiment of the invention, timing of the global period, and the counting are reset each said global period.

[0022] In an embodiment of the invention, the water channel is either a cold or a hot water channel.

[0023] In an embodiment of the invention, the conversion is made by an FFT converter.

[0024] The invention also relates to a system for detecting leakage in a domestic water channel, comprising: (a) a pressure sensor for acquiring a pressure signal S from the water channel; (b) a processing unit which is configured to: (i) receive said pressure signal S and record the same; (ii) sample the signal at a sampling rate s within a check period T; (iii) slide a sliding window W having a duration t over the check period T to acquire, in each position of W, a collection of N time-domain vectors Vt from signal S within period T; (iv) for each possible pair m,n within the N time-domain collected vectors Vt, calculate a distance value Dm,n; (v) given the plurality of values Dm,n, assigning each value Dm,n to a respective pixel within a synthetic image I; and (vi) feed the synthetic image I into a flow versus no-flow classifier to determine whether the synthetic image I is an Inf synthetic image indicating a no flow state, or an If synthetic image indicating a flow state of the water channel.

[0025] In an embodiment of the invention, the system further comprising a FFT converter configured to convert said N time-domain vectors Vt to frequency domain vectors Vf and wherein the calculations of the respective Dm,n vectors are calculated between pairs of frequency-domain vectors Vf. In an embodiment of the invention, the processing unit repeats steps (i)-(vi).

[0026] In an embodiment of the invention, the flow versus no-flow classifier is a machinelearning, neural-based classifier.

[0027] In an embodiment of the invention, the system further comprising a no-flow-state counter configured to count sequential determined no-flow synthetic images, and alert for a leakage if the number of sequential no-flow-synthetic images does not reach a threshold R at least Q times within a given global period.

[0028] In an embodiment of the invention, the slide is a step-by-step slide between samples.

[0029] Brief Description of the Drawings

[0030] In the drawings:

[0031] Fig. 1 schematically illustrates a typical water network of an apartment;

[0032] Fig. 2 is a block diagram describing the general structure of the leakage detection system of the invention;

[0033] Fig. 3 shows the system of the invention connected within a section of a cold water channel of the apartment;

[0034] Fig. 4a shows an example of pressure signal variations due to open / close events or during a steady state, according to the prior art;

[0035] Fig.4b shows various pressure events measured by the inventor within a site; Fig. 5a is a flow diagram illustrating a process for creating a representative pressure-variation synthetic image used as an input to the image-type classification machine learning unit, according to a first embodiment of the invention;

[0036] Fig. 5b is a time diagram demonstrating how the sliding window advances step by step within a selected section of a sampled pressure signal S.

[0037] Fig. 6a generally shows the construction of mage I from a given list of vectors Vf according to an embodiment of the invention;

[0038] Fig. 6b shows an example of a flow synthetic image I, as obtained in an experiment and following the process of Fig. 5a; Fig. 6c shows additional flow synthetic images I, as obtained in an experiment and following the process of Fig. 5a;

[0039] Fig. 6d shows an example of a no-flow synthetic image I as constructed in a steady-state case where there was no water consumption at all;

[0040] Fig. 6e shows additional no-flow synthetic images I, as obtained in an experiment and following the process of Fig. 5a;

[0041] Fig. 7 illustrates a process for determining a leakage, given a collection of runtime synthetic images I; and

[0042] Fig. 8 is a flow diagram showing process 800, in which the analysis is performed in the time domain.

[0043] Detailed Description of Preferred Embodiments

[0044] Fig. 1 schematically illustrates a typical water network of apartment 200, which includes a kitchen 230, a first bathroom 210, a second bathroom 220, and a laundry machine 270. In this example, several water-consuming devices are in the kitchen, such as a sink 232 and a dishwasher 234. The 1stbathroom 210 includes 1stbathtub sink 212, a first toilet 214, and a shower 216. The second bathroom 220 includes a 2ndbathtub sink 222, a 3rdbathtub sink 224, a 2ndtoilet 226, and a bathtub 228.

[0045] The cold water supply enters apartment 200 via inlet 280 and main valve 202, creating a cold water channel 240. After passing the main valve 202, pipe 242 of the cold channel 240 splits. Some water flows into heater 260, creating a hot water channel 250 at the heater’s outlet. Given this structure, water system 200 includes two separate (and isolated in terms of temperature) channels, a cold water channel 240 and a hot water channel 250, conveying hot and cold water to all the home’s water-consuming devices.

[0046] A water leakage may occur within the cold water channel 240 or the hot water channel 250. The leakage may occur within the channels’ infrastructure (namely, pipes, connectors, etc.) or within one or more of the water-consuming devices. The toilets and faucets are the two devices most prone to leakage (however, leakage may occur elsewhere within the water network). Fig. 2 is a block diagram describing the general structure of the leakage detection system 300 of the invention. Connector 302 is connected serially to the water flow within one of the cold or hot (preferably the cold) water pipes 240 or 250, respectively. Pressure sensor 310, being in water communication with the water flowing within connector 302, continuously senses the water pressure within the connector. The pressure measurement, which is representative of the pressure within the water channel (240 or 250, whichever is the case), is converted to an electric pressure signal 308 that is conveyed to processing unit 320 and recorded there for further processing and analysis, as elaborated hereinafter. When the processing unit’s analysis detects a water leakage, an alert signal 330 is generated. The processing unit 320 utilizes an image-type classification machine-learning unit 324 (also referred to as “classifier”) to determine leakage. The processing unit also includes a no-flow counter, whose function is described later below.

[0047] Fig. 3 shows the system 300 (of Fig. 2) connected within a section of cold water channel 240, in this case, within a pipe leading water to the 1stbathtub sink 212. Connector 302 may be connected anywhere within the cold or hot water channels and optionally utilized by flexible pipes and suitable connectors.

[0048] The analysis by the processing unit 320 is based on a machine-learning image classification structure, typically an image classification neural network. The system includes a pressure sensor 310 that continuously (24 hours a day) senses the pressure variations within the channel / s, mainly occurring due to intentional water consumption. When a leakage (unintentional consumption) occurs, the system detects this abnormal consumption and alerts.

[0049] The art has already acknowledged that the pressure within the water channels is correlated to the opening or closure of a water-consuming device. The steady-state pressure is supposed to be flat over time (sometimes with low intermittent amplitude fluctuations resulting from a pressure regulation mechanism applied by the local water distributor) when there is no leakage or other water consumption. Leakage and normal consumption introduce frequency components on the pressure signal, which, if appropriately observed, can distinguish between a water flow (leakage or any level of water consumption) and no flow (no leakage or consumption). Generally, the higher the consumption at a given time, the lower the frequency of the pressure variations on the signal. A minute flow, for example, due to leakage, introduces high-frequency components, while a higher level of water flow introduces lower-frequency components. Fig. 4a shows an example of pressure signal variations due to open / close events or during a steady state, according to US 11,493,371. Fig. 4b similarly shows various pressure events measured by the inventor within a site.

[0050] The invention utilizes an image-type classification machine-learning unit to detect leakage. Fig.5a is a flow diagram illustrating a process for creating a representative pressure-variation synthetic image used as an input to the image-type classification machine learning unit (324 in Fig. 2). The phrase “synthetic image” refers to an image that does not exist in reality, but is formed as a result of a process involving a plurality of other images. Moreover, the process of Fig. 5a provides an indication for a flow or no-flow state. For clarity, Fig. 5a is explained by reference to a specific non-binding example shown in Fig. 5b. Fig. 5b is a time diagram demonstrating how the sliding window advances step by step within a selected section of a sampled pressure signal S (the pressure signal S is not shown in the figure, but is represented by the samples). In step 402, a pressure signal S from one of the water channels 240 or 250 is recorded. In step 404, a checking period T (for example, 7 seconds - see also Fig. 5b) from signal S and a sampling rate s (for example, 50 samples per second) are defined. In step 406, a sliding window W having duration t (for example, 2 seconds = 100 samples’ duration) is defined. During each position of the sliding window W, t x s samples are acquired and recorded as as vector Vi. In step 408, the sliding window W starts “moving” along the samples s of signal S, starting from sample 1 and until position T-t (measured from the trailing end of W - in the example of Fig. 5b T-t = 5 seconds, i.e., at sample 250). It should be noted that the window “slides” in sample steps, for example, one sample at a time (samples 1-101, next samples 2-102, etc...), two samples at a time, etc. In the first position of window W, tx s samples are recorded as vector Vi (in the example of Fig.5b, 2 seconds x 50 samples per second = 100 recorded samples). Then, window W moves (for example) one sample, and the next collection of t x s samples is recorded as vector K?. The procedure of step 408 continues until the entire period from sample 1 to T-t is fulfilled, obtaining in total N time-domain vectors (in this specific example N=250 of vectors V1-V250 between sample 1 in which Vi is recorded until sample 250 in which V250 is recorded). Therefore, in this specific example N=250. If, in another example, the window W moves 2 samples at a time, the value of N is reduced to 125 in this specific example of Fig. 5b. Any level of the sliding window displacement, other than the one or two sample displacement discussed herein, is applicable. All the vectors V1-V250 in this example) are time domain vectors, also indicated as Vt. In step 410, an FFT (Fast Fourier Transform) is applied to each time-domain vector Vt obtained in step 408, transforming it to a frequency-domain vector Vf. In the present example, the result of this step is 250 frequency-domain vectors Vf. In step 412, a respective vector distance value D is calculated for each possible pair of vectors Vf. For example, this distance calculation includes the values D1,2,... D1,3, ... D1,4..., D1,n,... D3,n... Dn,n. Of course, the distance values Di,i, D2,2,... Dn,n are all zero, as each reflects a distance between a specific frequency domain vector and itself. The type of distance calculation may be selected from various distance calculation techniques known in the art, such as absolute, Euclidean, and square distance metrics, and many others, all within the scope of the invention.

[0051] It should be noted that the pressures at the cold and hot water channels are correlated. Therefore, while connecting the detection system 300 of the invention within the cold water channel 240 is preferable, this connection can still detect leakage occurring within both the hot and cold water channels 250 and 240, respectively. Similarly, the connection of the system within the hot water channel 250 can still detect leakage within both the hot and cold water channels 250 and 240, respectively.

[0052] Fig. 6a generally shows the construction of mage I from a given list of vectors Vf as obtained between the first sliding window W position (in the current example, spanning samples 1-100) and the last window position (in the example, samples 250-350). Each specific distance scalar value of a calculated distance vector D is assigned respectively within a pixel of synthetic image I, as shown. Fig. 6b shows an example of a 250X250 pixels synthetic image I 640, as obtained in an experiment and following the process of Fig. 5a, in one water operational state of a specific site, with a simulated leakage. As can be seen, the synthetic image is symmetrical about the diagonal line 610. This symmetry is expected, as the distance Dn,m always equals distance Dm,n for any specific n and m values. Also, in this particular example, each synthetic image I is a 250X250 pixels image. Various patterns have been observed for multiple water flow cases (including different types of leakage), as shown, for example, in Fig. 6c.

[0053] A 2D synthetic image is the preferred form for use in conjunction with the invention. Higher-dimensional synthetic images, e.g., 3D or 4D, may be used. Therefore, whenever the description and claims use “2D”, they should be interpreted as covering a higher dimension as well.

[0054] Fig. 6d shows an example of a synthetic image I as constructed in a steady-state case where there was no water consumption at all (either due to intentional consumption or due to leakage). Fig. 6e shows various other synthetic images I as constructed in additional steady-state cases where there was no water consumption or leakage.

[0055] As can be clearly seen, all the synthetic images If relating to various flow (even minute) cases, such as the If synthetic images of Figs. 6b and 6c include clear patterns. To the contrary, the no-flow synthetic images of Figs. 6d and 6e are monotonic, lacking any pattern. The flow synthetic images of Figs. 6b and 6c are well visually distinguishable from the no-flow synthetic images of Figs. 6d and 6e. The inventor also found that the transformation into the frequency domain (using the FFT to form Df images) significantly improves the visual distinguishment between a flow state and a no-flow state, compared to a similar synthetic image comparison performed on the time domain Dt images.

[0056] According to the invention, the image-type classification machine-learning unit 324 (Fig. 2), preferably in the form of a neural network, is trained to distinguish between a flow synthetic image If (i.e., an image reflecting one of many flow states) and a no-flow synthetic image Inf i.e., an image reflecting a no flow, and no leakage state). According to the invention, the process of Fig. 5a is repeatedly performed, e.g., 200 times per hour (or between 100 and 300 times per hour), indicating each time whether a flow or a no-flow state has been detected.

[0057] If there is no water leakage at the site, it can be assumed that a sequence of at least R sequential (or not sequential) no-flow synthetic images will appear, for example, during a period of 24 hours. R can be defined or tuned according to the preference or may be converted to a period Pi, for example, 1, 2, or 4 hours, in which only noflow synthetic images are detected. If, however, during a period P2 of, e.g., 24 hours, process 400 does not detect any repeated no-flow synthetic images to a threshold R (or a period Pi of no flow), system 300 assumes that there is a leakage and it generates an alert 330 (Fig. 2). During counting Inf to R, the system may “forgive” and ignore a few flow synthetic images If that may appear within the sequence due to noise. Such a forgiveness level may be predefined.

[0058] Fig. 7 illustrates a process 700 for determining a leakage, given a collection of runtime synthetic images I. In step 702, a threshold R of number of sequential no-flow synthetic images Inf is defined. As noted, if a sequence R (for example, of 10 or more consecutive Inf synthetic images) is not reached (once, twice, or any predefined number n of times), during a global predefined period P of, e.g., 1, 2, 4.... or 24 hours, a leakage alert is generated at the end of the process. In step 704, a run-time synthetic image I, resulting from process 400, is fed into the classifier 324. In step 706, the process determines whether the currently fed synthetic image I is a nonflow synthetic image Inf. In the negative case (i.e., determination of a flow synthetic image If), the Inf counter (i.e., a counter 326 shown in Fig. 2 that counts Inf synthetic images) is reset in step 720, and a subsequent synthetic image is fed in step 704. If, however, it is determined in step 706 that the synthetic image is an Inf (reflecting a no-flow), one count is added to the Inf (no-flow) counter 326 (Fig. 2) in step 708. In step 710, the content of the Inf counter is checked to determine whether the threshold R has been reached. If not, the procedure returns to step 704. Each time that R is reached, counter R (not shown in Fig. 2) is incremented in step 712. If threshold R is not reached in step 710, the procedure returns to step 704. In step 714, after the elapse of the global predefined period P of, e.g., 1, 2, 4.... or 24 hours, the procedure continues to step 716 in which the content of counter R is checked. If R ≥ 1, meaning that the occurrence of R consecutive Inf synthetic images has been detected at least once during the global period, the procedure concludes that there is no leakage. The counter R and the global timer (not shown in Fig. 2) are reset, and the procedure returns to step 704. Otherwise, if the answer in step 716 is negative, meaning that there was no single detection of R consecutive Inf synthetic images, the procedure concludes that there is a leakage, and a leakage alert is issued in step 730. It should be noted that in step 716, an alternative condition R ≥ Q may be defined, where Q indicates the minimal number that counter R is expected to be at the end of the global period before concluding that there is no leakage.

[0059] As noted above, procedure 400 (Fig. 5a) includes step 410 of transforming the time-domain vectors Vtto frequency-domain vectors Vf. As also noted, the inventor has found that performing the analysis in the frequency domain is preferable for distinguishing between flow synthetic images If and no-flow synthetic images Inf. However, the transformation to the frequency domain and the analysis in frequency-domain-related synthetic images is not the only option. Fig. 8 is a flow diagram showing process 800, in which the analysis is performed in the time domain. Procedure 800 (in which similar references to those of procedure 400 relate to similar functionalities) differs from procedure 400 in the following: (a) Step 410 is eliminated, and there is no equivalence to it in procedure 800; (b) In step 812, the distance calculation is performed on vectors Vt, rather than on vectors Vf (as in procedure 400). The rest of procedure 800 is substantially the same as procedure 400.

[0060] As noted, the image-type classification machine-learning unit 324 is preferably a neural network. This classification machine learning unit may be based on a computer vision algorithm, such as VGG16, Inception V3, ResNet50, etc. Example

[0061] The inventor has connected a pressure sensor to a water network within an apartment and acquired pressure signals such as those shown in Fig. 4b. The apartment included a water network as shown in Fig. 1. The inventor has applied the method of Figs. 5a and 7, where the period T was 7 sec, the sampling rate s was 20ms (i.e., 50 samples per second), with a sliding window W having duration t of 2sec, with a striding size of 40ms. About 50 experiments were conducted during water consumption (flow), and at various levels of no-consumption (no-flow). Based on the vectors Vt obtained during each experiment, various synthetic images were created either according to the frequency domain process of Fig. 5a or according to the time domain process of Fig. 8. Following the creation of these synthetic images, the inventor trained a computer vision (CV) algorithm model using VGG16 (a convolutional neural network architecture developed by Simonyan et al.) using labeled synthetic images If or Inf. The CV model was trained to classify two possible classes, flow or no-flow. A counter was used to count the number of hours the water usage occurred without no-flow detections. Then, in run-time, the flow counter was reset each time there was a sequence of 25 no-flow detections. When the counter was not reset during a continuous period of 12 hours, it was concluded that a leak existed in the system, and an alert was generated. Various types of leakages were simulated, and in all the cases, the leakage was detected. The process of Fig. 5a was found to be superior to the process of Fig. 8.

[0062] While some embodiments of the invention have been described by way of illustration, it will be apparent that the invention can be carried into practice with many modifications, variations, and adaptations and with the use of numerous equivalent or alternative solutions that are within the scope of persons skilled in the art, without departing from the spirit of the invention or exceeding the scope of the claims.

Claims

Claims1. A method for detecting leakage in a domestic water channel, comprising:(a) acquiring by a single sensor positioned at a single location along the water channel and recording a pressure signal S from the water channel; (b) defining a check period T and a sampling rate s, and sampling the signal S during the period T, in the sampling rate s;(c) defining a sliding window W having a duration t;(d) activating sliding of window W to slide within period T and to acquire, in each position of W, a collection of N time-domain vectors Vt from signal S;(e) for each possible pair m,n within the collected vectors Vt, calculating a respective distance value Dm,n;(f) given the plurality of values Dm,n, assigning each value Dm,n to a respective pixel within a synthetic image I; and(g) feeding the synthetic image I into a flow versus no-flow classifier to determine whether the synthetic image I is an Inf synthetic image indicating a no flow state, or an If synthetic image indicating a flow state of the water channel.

2. The method of claim 1, wherein each of the N time-domain vectors Vt is converted to a frequency domain vector Vf and wherein the calculations of the respective D vectors are calculated between each possible pair m,n of the frequency-domain vectors Vf.

3. The method of claim 1, further comprising repeating steps (a) and (d)-(g), thereby to repeatedly determine no flow state synthetic images Inf or flow state synthetic images If within the channel.

4. The method of claim 1, wherein the flow versus no-flow classifier is a machine-learning, neural-based classifier.

5. The method of claim 1, further counting sequential no-flow synthetic images, and alerting for a leakage if the counted sequential number of no-flow-synthetic images does not reach within a predefined global period a threshold number R at least a predefined number of times Q.

6. The method of claim 5, wherein said given global period is in the order of at least 4 hours, and wherein the number of occurrences Q depends on said given global period.

7. The method of claim 6, wherein the given global period, the threshold number R, and the number of occurrences Q are tunable by a user.

8. The method of claim 5, wherein said counting is reset each said global period, and / or each occurrence of a flow synthetic image If.

9. The method of claim 1, wherein said water channel is either a cold or a hot water channel.

10. The method of claim 2, wherein the conversion is made by an FFT converter.

11. A system for detecting leakage in a domestic water channel, comprising:(a) a single pressure sensor positioned at a single location along the water channel for acquiring a pressure signal S from the water channel;(b) a processing unit which is configured to:(i) receive said pressure signal S and record the same;(ii) sample the signal at a sampling rate s within a check period T;(iii) slide a sliding window W having a duration t over the check period T to acquire, in each position of W, a collection of N time-domain vectors Vt from signal S within period T;(iv) for each possible pair m,n within the N time-domain collected vectors Vt, calculate a distance value Dm,n;(v) given the plurality of values Dm,n, assigning each value Dm,n to a respective pixel within an synthetic image I; and(vi) feed the synthetic image I into a flow versus no-flow classifier to determine whether the synthetic image I is an Inf synthetic image indicating a no flow state, or an If synthetic image indicating a flow state of the water channel.

12. The system of claim 11, further comprising a FFT converter configured to convert said N time-domain vectors Vt to frequency domain vectors Vf and wherein the calculations of the respective Dm,n vectors are calculated between pairs of frequency-domain vectors Vf.

13. The system of claim 11, wherein the processing unit repeats steps (i)-(vi).

14. The system of claim 11, wherein the flow versus no-flow classifier is a machine-learning, neural-based classifier.

15. The system of claim 11, further comprising a no-flow-state counter configured to count sequential determined no-flow synthetic images, and alert for a leakage if the number of sequential no-flow-synthetic images does not reach a threshold R at least Q times within a given global period.

16. The system of claim 11, wherein the slide is a step-by-step slide between samples.

Images