METHOD AND PORTABLE DEVICE FOR DETECTING LEAKS IN A PIPELINE

Description

TECHNICAL FIELD

[0001] The present invention belongs to the field of signal processing devices, in particular of acoustic signals emitted during a fluid-structure interaction, and relates more particularly to a method of leak detection in a pipeline carrying any fluid and a portable device for implementing such a method.

[0002] The present invention finds a direct, but not exclusive, application in the detection of leaks in a drinking water network. STATE OF THE ART

[0003] Pipes, especially pressurized supply pipes such as water lines in a drinking water network or industrial fluid lines in factories, are susceptible to various failures that can cause leaks.

[0004] These leaks compromise the normal functioning of the pipelines and can have serious technical, economic and environmental consequences.

[0005] Detecting these leaks often proves to be a difficult task, particularly because of the difficulty of accessing the pipes, which are sometimes buried, and their size.

[0006] Some "artisanal" solutions rely on human analysis of the sound emitted by a pipe and recorded with a microphone or probed with a stethoscope.

[0007] These solutions are tedious, imprecise, and require the intervention of highly skilled technicians.

[0008] US4309576 describes a listening device for locating water leaks, comprising a manually operated rod with a handle at one end for manually positioning the listening device. An acoustic pickup is mounted at the other end of the rod, and it includes a ceramic audio tone transducer that is operationally connected to an amplifier-receiver with a level indicator on which visual leak signals can be seen. The acoustic pickup is also connected to headphones to allow the leaks to be heard. The transducer includes a brass diaphragm directly attached to a threaded stud for mounting in the rod.

[0009] There are so-called invasive solutions because they require the introduction of an object into the pipe to detect the leak.

[0010] For example, we know of methods using photonic detection technology that transforms a fiber optic cable running along a network of water pipes into thousands of vibration sensors, capable of detecting a disturbance along the entire length of the pipe.

[0011] Document FR2935800 describes a method for detecting leaks in an underground liquid pipe, in particular a water pipe, according to which: a gas is injected, via a diffuser, into the liquid of the pipe, a gas whose concentration in the atmosphere is low, and the path of the pipe is traversed on the surface with a detection system to measure at successive points the concentration of the injected gas in the air, an abnormally high concentration constituting an indication of leakage.

[0012] This solution is clearly extremely expensive and very impractical.

[0013] Document EP2710291 describes a system for identifying leaks in liquid piping systems, comprising at least one detection unit positioned near at least one outlet point of said piping system and including an acoustic sensor and a wireless communication unit; an electronically controlled shut-off unit installed at an inlet point of the piping system, the unit including a valve, an acoustic sensor, and a wireless communication unit arranged to transmit water flow data and receive control signals; and a controller network device for receiving measurement data from all sensors. The controller is programmed to detect leaks when it identifies differences between the liquid flow measured at the inlet point and the flow measured at the outlet points.

[0014] Document WO2004 / 063623 describes a method and device for detecting potential leaks in a pipeline. The pipeline is continuously monitored by acoustic monitoring devices, and acoustic events indicating a potential leak are recorded. The pipeline is also equipped with means for continuous temperature monitoring, either periodically or on demand. According to this solution, a leak is considered probable when an acoustic event indicates a possible leak at a location and if, at approximately the same time, a temperature difference exceeding a predetermined value is detected between that location and adjacent locations. In this solution, the temperature can be monitored by a satellite, aircraft, or drone.

[0015] Document EP2028471 describes a leak detector that provides accurate and stable detection of the presence and location of a leak in an underground water pipe. The leak detector includes a vibration detector with a sensor incorporating a piezoelectric element, a main detector body incorporating voltage amplifiers to amplify the output signal voltage, several types of noise reduction units to eliminate noise from the output signal, and a receiver. The main detector body has a display unit to show detected sound data on a predefined screen.

[0016] These prior art solutions are complex to implement and do not allow for rapid and highly accurate leak detection. Furthermore, none of these solutions is a portable, user-friendly, and compact device to facilitate the intervention of leak detection technicians.

[0017] Document AU2020262969A1 describes a signal detection system to identify structural anomalies in a pipeline network, but does not disclose the extraction of the DSF (defect-sensitive characteristic) as well as the extraction of the center of gravity of the periodogram, nor the principal component analysis step to model a normal signal of the pipeline and detect a possible leak by comparing the observed signal to the normal signal.

[0018] The document “CLARK CASEY ET A: "Wireless leak detection using airborne ultrasonics and a fast-Bayesian tree search algorithm with technology demonstration on the ISS" 2015 IEEE International Conference on Wireless for Space and Extreme Environments (WiSEE), IEEE, December 14, 2015 » This document relates to a leak detection method in a pipeline that uses the periodogram and is implemented by a portable device including a microphone. It does not disclose the DSF extraction or PCA analysis.

[0019] The document « Toshitaka Sato, Akira Mita: "Leak detection using the pattern of sound signals in water supply systems", Proceedings volume 6529, sensors and smart structures technologies for Civil, Mechanical, and Aerospace systems 2007, Vol. 6529, April 6, 2007 »relates to a method for detecting leaks in a pipeline which relates to the methods of the PCA model and DSF, but does not disclose the extraction of features (center of gravity and entropy) and the combination of the results of the extracted features (entropy, DSF, and center of gravity) and the PCA analysis. PRESENTATION OF THE INVENTION

[0020] The present invention aims to overcome the disadvantages of the prior art described above and offers a simple solution for use allowing automatic leak detection in a pipeline in a very short time (less than one minute).

[0021] A main objective of the invention is to provide a portable device that can instantly discriminate the presence or absence of a leak in a pipe with very high accuracy, reaching 100% in most cases.

[0022] Another objective of the invention is to use a set of artificial intelligences to optimize leak detection and make it discreet and fast, and therefore implementable in a laptop or a smartphone-type mobile phone.

[0023] To this end, the present invention relates to a method in accordance with claim 1 for the electro-acoustic detection of a leak in a pipeline carrying a fluid, comprising: a step of acquiring a sound signal emitted by the pipeline using a microphone; and a step of real-time signal processing.

[0024] For a better result, the signal entropy is calculated using an approximate entropy model ApEn ( Approximate Entropy ).

[0025] Advantageously, the DSF characteristic is calculated from an AR (autoregressive) model of the signal.

[0026] More specifically, the AR model of the signal is an ARIMA-adjusted mean integrated model ( Auto-Regressive Integrated Moving Average ) presenting ARIMA coefficients, and the DSF characteristic is equal to the ratio between the first ARIMA coefficient and the square root of the sum of the squares of the first three ARIMA coefficients.

[0027] In an advantageous embodiment, the PCA analysis step of the signal implements artificial intelligence algorithms, such as neural networks, to improve leakage detection by assigning a weight to each feature calculated in the feature extraction step. It should be noted that the weights and coefficients used in the preceding calculations are optimized by a genetic algorithm for each application after a training period.

[0028] The present invention also relates to a portable device comprising a microphone and computing means in the form of a microprocessor, for implementing a leak detection method in a pipeline as described. Advantageously, the portable device includes a mobile terminal such as a smartphone connected to the microphone, the method then being executed in a dedicated mobile application.

[0029] The fundamental concepts of the invention having been set forth above in their most elementary form, other details and characteristics will become clearer from the reading of the following description and with regard to the attached drawings, giving by way of non-limiting example an embodiment of a method and a device for detecting leaks in a pipeline, in accordance with the principles of the invention. BRIEF DESCRIPTION OF THE FIGURES

[0030] The figures are provided for illustrative purposes only to aid understanding of the invention and do not limit its scope. The various elements are represented schematically and are not necessarily to the same scale. Throughout the figures, identical or equivalent elements are assigned the same numerical reference.

[0031] This is illustrated as follows: Figure 1 : a portable leak detection device placed near a pipeline, according to a first embodiment of the invention; Figure 2 : an flowchart of the main steps of a leak detection process in a pipeline according to the invention; Figure 3 : a portable leak detection device, according to a second embodiment of the invention. DETAILED DESCRIPTION OF IMPLEMENTATION METHODS

[0032] The embodiment described below refers to a portable leak detection device for pipes, primarily intended for locating leaks in water pipes. This non-limiting example is given for a better understanding of the invention and does not preclude the use of the device for detecting leaks in other pipes such as pipelines, gas pipelines, oil pipelines, and more generally, all industrial fluid circulation systems in factories or power plants.

[0033] In this description, the term "portable device" refers to a small device that a user can carry in their hands for use, like a mobile phone.

[0034] There figure 1 represents a portable leak detection device 100 placed near a pipe 200 in which a fluid F flows in a normal direction of flow (in thick arrows).

[0035] Pipeline 200 is subject to a leak L (in thin arrow) caused in particular by damage such as a crack 210.

[0036] Naturally, the leak L induces a particular acoustic signature in the sound signal S produced by the fluid flowing in the pipe 200.

[0037] The portable device 100 thus makes it possible to probe the sound signal S emitted by the pipe 200 and to detect the presence of the leak L by recognizing its acoustic signature.

[0038] Prior to this recognition, a microphone integrated or connected to the portable device 100 allows acquisition of the sound signal S when it is placed near the pipe 200.

[0039] Of course, the recognition of the acoustic signature is the result of a specific signal processing carried out by the computing means of the portable device 100.

[0040] Thus, the portable device 100, thanks to its microphone and its onboard computing means, implements a leak detection process according to the present invention.

[0041] There figure 2 represents the main steps of a 500 leak detection process in a pipeline, said process comprising: a step 510 of acquiring the sound signal emitted by the pipeline; a step 520 of processing this signal according to a sequence of well-defined operations; a step 530 of applying a Kalman filter; a step 540 of principal component analysis; a step 550 of verifying the presence of a leak; and a conditional step 560 of alerting and / or notifying in case of leak detection.

[0042] Step 510 of sound signal acquisition consists of recording this signal by the microphone of the portable device 100, at different accessible points of the pipeline.

[0043] The signal is then analyzed in a processing and computing unit of the portable device 100.

[0044] Signal processing step 520 is performed using a frequency-domain approach on the continuous signal, which is the sound signal from the pipeline. This involves real-time digital signal processing in an embedded microprocessor, preferably one specialized in digital signal processing.

[0045] The signal processing performed during step 520 includes: a step 521 of frequency analysis of the signal; and a step 522 of extraction of signal features;

[0046] Step 521 of frequency analysis consists of obtaining a frequency representation of the recorded signal using a periodogram ( periodogram ) in order to estimate its power spectral density (PSD).

[0047] DSP allows, within the framework of harmonic analysis, the characterization of the random and stationary signal that is the sound signal of the pipeline.

[0048] This makes it possible to identify the harmonics (or fundamental frequencies) of the sound signal, which therefore correspond to a normal operation of the pipe, and to remove external sounds from the environment which are parasitic sounds to the leak being sought.

[0049] Step 522 of signal feature extraction consists of calculating features from a previously established AR (autoregressive) model of the signal.

[0050] This feature extraction includes, in particular, a 5221 calculation of entropy, a 5222 calculation of a fault-sensitive characteristic called DSF ( Damage-Sensitive Feature ) and a calculation of 5223 of the centered frequency ( centered frequency ) which is the barycenter of the periodogram.

[0051] Indeed, as soon as there is an anomaly in the operation of the pipe and this anomaly alters the resulting sound signal, there is necessarily more complexity in said signal and therefore more entropy. Thus, the entropy of the sound signal from a pipe with a leak is necessarily greater than that of the signal from the same pipe without a leak.

[0052] Preferably, the entropy of the recorded signal is calculated using the ApEn approximate entropy statistical model ( Approximate Entropy ). The ApEn model yielded better results compared to other entropy calculation models such as the SampEn sample entropy model ( Sample Entropy ), which is a model derived from ApEn, or Kolmogorov-Sinai entropy.

[0053] In addition, the ApEn model has many advantages, including low computational consumption, operation for small data samples, real-time operation, and limited noise effect on measurements.

[0054] The AR model of the signal is preferably an integrated autoregressive model with adjusted mean ARIMA ( Auto-Regressive Integrated Moving Average ) whose ARIMA coefficients allow the calculation of the DSF which is equal to the ratio of the first coefficient and the square root of the sum of the squares of the first three coefficients.

[0055] Thus, normal signals have frequencies concentrated in the same region, while abnormal signals (with leakage) do not have such a concentration of frequencies.

[0056] Then, based on all these characteristics, a principal component analysis (PCA) and a Kalman filter are applied to detect and locate any leaks in the pipeline.

[0057] Step 530 of applying a Kalman filter makes it possible to estimate the state of the system representing the flow in the pipeline, in order to generate the system residues, from a model of good system operation (without leakage) and the available measurements, these residues being the acoustic signatures revealing the presence of leakage in the pipeline.

[0058] Step 540 of PCA analysis allows the generation of a PCA model in which all correlations between the different characteristics are taken into account.

[0059] Thus, PCA analysis makes it possible to model the behavior of the pipeline (system) in normal operation, and leaks (defects) are then detected by comparing the behavior observed in the recorded sound signal and that given by the PCA model.

[0060] Combining the results of the features extracted in step 520 with PCA analysis yields a segmentation of a two-dimensional space onto which are projected the vectors of a higher-dimensional space (e.g., 5) based on the extracted features. This segmentation separates leak-free signals from leaky signals, thus enabling leak detection. Preferably, method 500 employs artificial intelligence algorithms such as neural networks to enhance PCA analysis by assigning a weight to each of the features calculated in step 520. It should be noted that the weights and coefficients used in the preceding calculations are optimized by a genetic algorithm for each application after a training period.

[0061] There figure 3represents a portable device 100', according to another embodiment, for implementing the leak detection method, said device comprising a detached microphone 10 and a mobile terminal such as a smartphone 20.

[0062] The microphone 10 and the smartphone 20 can be connected with a cable 30 or via a wireless link.

[0063] Thus, the portable leak detection device can simply be transported and used by a technician wishing to check the condition of a pipe.

[0064] It is clear from this description that certain steps of the process can be modified, replaced or deleted and that certain adjustments can be made to the implementation of this process according to the objectives pursued, without departing from the scope of the invention as defined by the claims.

Claims

1. Method (500) for electro-acoustic detection of a leak in a pipeline conveying a fluid, comprising: - a step (510) of acquiring a sound signal (S) emitted by the pipeline by means of a microphone (10); and - a step (520) of processing the signal (S) in real time; characterised in that the step (520) of processing the signal comprises: - a step (521) of frequency analysis of the signal (S) by means of a periodogram allowing the power spectral density (PSD) of said signal to be estimated; and - a step (522) of extracting features of the signal (S), including the entropy, a fault-sensitive feature referred to as DSF, and a centred frequency corresponding to the centroid of the periodogram; in that it comprises, after the above steps: - a step (530) of applying a Kalman filter to estimate residuals corresponding to signals indicative of a leak, said residuals being the acoustic signatures indicative of the presence of a leak in the pipeline; - a step (540) of principal component analysis allowing a normal signal of the pipeline to be modelled and a possible leak to be detected by comparing the observed signal (S) with the normal signal; and - a step (560) of presenting the results and of issuing an alert and / or a notification in the event of detection of a leak; and in that it is entirely implemented by a portable device (100, 100') comprising the microphone (10).

2. Method according to claim 1, wherein the entropy of the signal (S) is calculated using an approximate entropy model.

3. Method according to claim 1 or 2, wherein the DSF feature is calculated from an autoregressive model of the signal (S).

4. Method according to claim 3, wherein the autoregressive model of the signal (S) is an autoregressive integrated moving average model having coefficients referred to as ARIMA coefficients, and the DSF feature is equal to the ratio between the first ARIMA coefficient and the square root of the sum of the squares of the first three ARIMA coefficients.

5. Method according to any one of the preceding claims, wherein the step (540) of principal component analysis of the signal (S) implements artificial intelligence algorithms for improving leak detection by assigning a weight to each feature calculated in the feature extraction step (522).

6. Method according to claims 4 and 5, wherein the weights and the coefficients used in the calculations are optimised by a genetic algorithm for each implementation after a learning period.

7. Portable device (100, 100') comprising a microphone (10) and computing means in the form of a microprocessor, for implementing a method (500) of leak detection in a pipeline according to one of claims 1 to 5.

8. Portable device (100') according to claim 7, comprising a mobile terminal of the smartphone type (20) connected to the microphone (10), the method (500) being executed in a dedicated mobile application.