Pipeline fault identification system

The fault identification system in pipelines uses a robot and statistical learning model to accurately classify multiple defects in non-visitable pipes, enhancing precision and efficiency while reducing operator workload and maintaining compatibility with existing practices.

FR3122924B1Active Publication Date: 2024-11-15VEOLIA ENVIRONNEMENT
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
FR2021005052
Authority / Receiving Office
FR · FR
Patent Type
Patents
Current Assignee / Owner
Filing Date
2021-05-12
Publication Date
2024-11-15
Estimated Expiration
2041-05-12

AI Technical Summary

Technical Problem

Existing pipeline inspection methods face challenges in accurately and efficiently identifying multiple types of defects in non-visitable pipes, such as wastewater pipes, due to the complexity of manual classification, subjectivity, and the need for specialized equipment, which are not adequately addressed by current automated solutions.

Method used

A fault identification system using a robot equipped with a camera and a statistical learning model to analyze photographs for multiple defect types, providing precise defect classification and operator support, while maintaining compatibility with existing inspection practices.

Benefits of technology

Enhances the precision and efficiency of defect identification, reduces operator workload, and improves compliance with standards by enabling simultaneous classification of multiple defects in a single photograph, without requiring significant additional training or specialized equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000018_0000
    Figure 00000018_0000
  • Figure 00000018_0001
    Figure 00000018_0001
Patent Text Reader

Abstract

A pipe fault identification system to facilitate the determination of the type of faults present in such a pipe, and a method for training a statistical learning model for such use and a method for identifying faults in a pipe, the identification system comprising a robot (40), equipped with a locomotion device (41) and a camera (42), configured to travel along the pipe and take photographs of portions of the pipe suspected of having one or more faults, and a processing system (20), equipped with a statistical learning model (22) configured to determine, on the basis of such a photograph, the type of each fault identified in the photograph from a pre-established list of at least 10 types of faults. Fig. 1.
Need to check novelty before this filing date? Find Prior Art

Description

Title of the invention: System for identifying faults in a pipeline Technical field

[0001] The present disclosure relates to a system for identifying faults in a pipeline making it possible to facilitate the determination of the type of faults present in such a pipeline. It also relates to a method for training a statistical learning model intended for such use as well as a method for identifying faults in a pipeline.

[0002] Such an identification system can in particular be used in the context of the maintenance of a network of non-inspectable pipes, typically wastewater, in order to assess the state of health of the pipes, to improve the accuracy of the identification of defects and to facilitate the drafting of an inspection report. Prior art

[0003] Buried pipeline networks, particularly wastewater networks, are subject to regular inspections intended to identify any defects present in the pipeline with a view to planning possible maintenance of all or part of the network.

[0004] Since most of the pipes in these networks are not accessible for inspection, i.e. too small for an operator to enter the pipe, it is customary to use a robot equipped with a camera to scan the pipes in search of defects. An operator then controls the movement of the robot and views the video stream transmitted by the robot's camera in order to identify any defects in real time. When a defect is detected, the operator is responsible for taking a photograph of the section of pipe affected by the defect and determining the type of defect among the types of a classification defined by the NF-13508-2 standard: the operator thus draws up, as defects are discovered, an inspection report associating the photographs of the defects with their classification.

[0005] However, this standard includes a large number of defect categories (around thirty) as well as an even larger number of subcategories (several hundred). The classification work is therefore particularly demanding and requires operators with significant skill and know-how. It is also tedious since the operator must refer to this standard and its codification for each of the defects encountered. Furthermore, this manual identification work is sensitive to the subjectivity of the operator and to possible cognitive biases.

[0006] Attempts at automatic fault identification have then been proposed. However, to date, these attempts are essentially academic and do not corre do not respond to the reality on the ground. In particular, most of the solutions proposed to date focus on the characterization of a specific type of defect, for example cracks, but are unable to identify other types of defect, such as the presence of a deposit or roots for example.

[0007] Some more general solutions exist but they are nevertheless limited to a small number of types of defect, rarely more than a few units, and therefore do not offer sufficient precision to satisfy the level of precision required by these inspection reports. In addition, most of the solutions proposed to date are limited to the classification of a single defect and are incapable, in particular, of simultaneously classifying several defects when several defects are present on the same photograph, which happens frequently in practice, certain defects being able to favor the appearance of a second defect of another type.

[0008] In addition, most existing solutions propose using special cameras, such as fish-eye cameras, to produce a 3D reconstruction of the pipeline, including its defects. However, these solutions are not preferred because they require investing in specific equipment and training operators in a completely different way of working. They also cause significant distortion of the images.

[0009] There is therefore a real need for a system for identifying faults in a pipeline which is free, at least in part, from the drawbacks inherent in the aforementioned known solutions. Statement of the invention

[0010] The present disclosure relates to a system for identifying defects in a pipeline, comprising a robot, equipped with a locomotion device and a camera, configured to travel through the pipeline and take photographs of portions of the pipeline suspected of having one or more defects, and a processing system, provided with a statistical learning model configured to determine, on the basis of such a photograph, the type of each defect recognized in the photograph from a pre-established list of at least 10 types of defects.

[0011] Thanks to such a statistical learning model, it is possible to determine the type of defects present in the photographs or at least to have a sufficiently precise estimate of the probability of presence of each defect to provide the operator with decision-making assistance in identifying each defect, for example in the form of a classification proposal. The production of inspection reports is thus facilitated. The accuracy and rigor of the inspection reports are also improved.

[0012] In particular, since the statistical learning model is trained to re know a large number of different types of defects, compliance with the NF-13508-2 standard, when sought, is easier to achieve without significant additional effort on the part of the operator. Since the statistical learning model is also capable of simultaneously determining the type of several defects present on the same photograph, the operator is not forced to spend time trying to isolate each defect on a separate photograph: the inspection is therefore faster and the inspection report shorter, as a single photograph can describe several defects.

[0013] Furthermore, such an identification system is based on the conventional inspection operation as it is carried out to date, with the use of a robot traveling along the pipeline looking for defects, taking photographs of the portions of pipeline exhibiting these defects, and determining the type of defect on the basis of this photograph. The transition from the conventional operation to this operation assisted by the statistical learning model is therefore easy: in particular, previous robots can be used and the additional training of the operators is limited to that of handling the results of the statistical learning model.

[0014] In particular, as in the conventional inspection operation, the determination is carried out on the basis of single photographs, and not on the basis of a video or a time and / or spatial series of photographs of the same defect. This makes it possible in particular to place oneself in the same operational conditions as the past inspections and therefore to take advantage during the training of the statistical learning model of the large photographic database of past inspections available to the operator, thus ensuring satisfactory reliability of the statistical learning model from its first use.

[0015] In certain embodiments, the pipeline is a non-inspectable pipeline, having a nominal diameter of less than 1600 mm, preferably between 200 and 500 mm.

[0016] In some embodiments, the pipeline is a wastewater pipeline.

[0017] In some embodiments, the pre-established list comprises at least 15 types of defects, preferably at least 20 types of defects, more preferably at least 30 types of defects. The more the statistical learning model is able to identify different types of defects, the more the residual work of the operator is reduced, especially when it is necessary to comply with the codification of the standard.

[0018] In certain embodiments, all the types of defects in the pre-established list are part of the types of defects described by the NF-13508-2 standard.

[0019] In some embodiments, the pre-established list includes at least the following types of defects: crack (BAB); surface degradation (BAF); defective connection (BAH); visible ground (BAO); deposits and adherent deposits (BBC+BBB); rupture / collapse (BAC); penetrating connection (BAG); pipe displacement (BAJ); roots (BBA); and connection (BCA). The trigrams in parentheses correspond to the codifications of the NF-13508-2 standard.

[0020] In some embodiments, each photograph comprises at least one defect. This facilitates the work of determining the statistical learning model which can start from the hypothesis that at least one defect is present: the statistical learning model can thus concentrate on determining the type of defects, which reinforces its precision in this task. In particular, the statistical learning model can be trained for this purpose on a database containing exclusively photographs having at least one defect.

[0021] In some embodiments, at least some photographs include multiple defects.

[0022] In some embodiments, the robot is configured to be driven by an operator. However, in other embodiments, the robot could be configured to navigate the pipeline autonomously.

[0023] In some embodiments, the robot's camera is orientable. It is preferably oriented by the operator. In the context of autonomous control of the robot, it could be oriented autonomously.

[0024] In certain embodiments, the robot's camera is an RGB camera. It is therefore a conventional camera such as already exists on this type of robot.

[0025] In certain embodiments, the robot's camera is provided with a lens having a focal length greater than or equal to 20 mm, preferably greater than or equal to 25 mm, more preferably greater than or equal to 30 mm. It is therefore a conventional camera such as already exists on this type of robot. In particular, the robot's camera is not provided with a fish-eye lens. In this way, the photographs taken by the camera do not present any notable distortion, which facilitates the recognition work of a potential operator. In addition, it is possible to avoid disturbing the image recognition by the statistical learning model if its learning has been carried out on photographs without distortion.

[0026] In some embodiments, the photographs are triggered manually by an operator.

[0027] In some embodiments, the processing system comprises an automatic defect detection module configured to detect at least certain types of defects based on the video stream from the robot's camera. Such an automatic detection module is particularly advantageous in the context of autonomous control of the robot: the routing of the pipeline, the detection of defects, the orientation of the camera and the taking of relevant photographs can thus be done autonomously, without intervention from the operator. In this case, this detection module au- The automated defect analysis is preferably separate from the statistical learning model.

[0028] In some embodiments, the processing system is configured to receive photographs taken by the robot without delay and transmit them without delay to the statistical learning model for determining the types of defects. In other words, the determination of defects is done on the spot, as defects are located in the pipeline.

[0029] In other embodiments, the determination can be made later, on the basis of all the photographs recorded during the journey of the section.

[0030] In some embodiments, the identification system further comprises a database of photographs comprising at least 1,000, preferably at least 100,000, more preferably at least 1,000,000, photographs of pipeline portions having at least one defect, each photograph being associated with the type of each of the defects visible in the photograph. The larger this database, the more extensive the training of the statistical learning model will be and therefore the more accurate it will be. Furthermore, as mentioned above, it is preferable to only include photographs having at least one defect in order to specialize the statistical learning model in the classification of defects rather than in their simple detection.

[0031] In some embodiments, the statistical learning model has been trained on said database.

[0032] In some embodiments, the statistical learning model is a multi-label classifier. Unlike a multi-class classifier that predicts multiple but mutually exclusive classes, a multi-label classifier predicts multiple and non-mutually exclusive classes: in other words, based on a given photograph, such a multi-label classifier is capable of determining the type of several defects belonging to different classes.

[0033] In some embodiments, the statistical learning model comprises at least one neural network, preferably of the deep convolutional type.

[0034] In certain embodiments, the statistical learning model comprises a hybrid neural network, assembling convolutional and recurrent architectures. Such a hybrid architecture makes it possible in particular to take advantage of correlations which may exist between certain types of defects, certain defects being able to promote the appearance of a second defect of another type.

[0035] In some embodiments, the statistical learning model comprises several classifiers trained on different databases. It is thus possible to combine the predictions of these models in order to improve the performance and robustness of the predictions.

[0036] In some embodiments, the learning model is configured to de complete a probability of presence for each of the types of defects on the pre-established list.

[0037] In some embodiments, the learning model is configured to determine that a particular type of defect is present if it determines that the probability of presence of this type of defect is greater than a predetermined threshold. Adjusting this threshold makes it possible to find a compromise between suggesting a large number of defects, at the risk of generating a significant number of false positives that the operator will have to correct manually to the detriment of the speed of the inspection, and suggesting a smaller number of defects, with a lower probability of false positives but a greater risk of false negatives and therefore to the detriment of the quality of the inspection.

[0038] In some embodiments, each type of fault is associated with a personalized threshold. Indeed, it may be advantageous to offer the operator a greater proportion of false positives for certain types of fault which are frequently omitted by operators or which are critical for operation and a greater proportion of false negatives for other types of fault which the operator can easily detect and correct by himself.

[0039] In some embodiments, the identification system further comprises a human-machine interface comprising at least one screen, the processing system being configured to display on the screen the current photograph accompanied by a classification proposal for each of the defects recognizable on the photograph. The identification system thus provides decision support to the operator who remains in control of the final classification. This speeds up the inspection work while ensuring a very high level of reliability, any errors in determining the statistical learning model being able to be corrected by the operator.

[0040] In some embodiments, the processing system is configured to further display the position of the recognized defects superimposed on the photograph. Implementing an algorithm for visualizing the activation maps is a means of providing an index of the position of the defects in the image, by highlighting areas of the image likely to contain the defects.

[0041] In some embodiments, the processing system is configured to receive a confirmation or correction of the proposal, entered by an operator via the human-machine interface, and to record the final classification for each of the defects in the photograph. The correction may include correcting the type proposed for a given defect and / or adding an additional defect not recognized by the statistical learning model with its classification as determined by the operator.

[0042] In some embodiments, the processing system is configured to transmit the photograph and the classification of each of the defects in the photograph to a report writing software.

[0043] In some embodiments, the processing system is configured to add each new photograph to the database, in association with the final classification of defects. This addition is preferably carried out after validation or correction by the operator. This makes it possible to enrich the database with each new inspection.

[0044] In some embodiments, the processing system is configured to evaluate the statistical learning model based on confirmations or corrections entered by the operator.

[0045] In some embodiments, the processing system is configured to adjust the parameterization of the statistical learning model based on the confirmations or corrections entered by the operator. This adjustment can also take into account the prediction scores of the model. The statistical learning model therefore continues to learn thanks to the corrections entered by the operator, which improves its accuracy for future inspections. Such an adjustment can in particular be carried out at regular intervals or when too significant a drift is detected. Preferably, such an adjustment can be made outside the inspection phases, that is to say after the robot has completed its journey through the pipeline.

[0046] In some embodiments, the processing system takes the form of a central processing unit. However, the processing system could also be at least partially delocalized: for example, the database and / or the statistical learning model could be hosted on a remote server, for example within a computing cloud.

[0047] In certain embodiments, the identification system is on board a vehicle, for example a utility vehicle such as a van. The identification system can therefore be brought as close as possible to the pipeline to be inspected in order to identify defects, classify them and write the report on site, during the robot's journey along the pipeline.

[0048] The present disclosure also relates to a method for training a statistical learning model intended for identifying faults in a pipeline, comprising constructing a master database of photographs comprising at least 1,000, preferably at least 100,000, more preferably at least 1,000,000, photographs of pipeline sections exhibiting at least one defect, each photograph being associated with the type of each of the defects visible in the photograph, and training the statistical learning model on the database main one thus constructed.

[0049] Such a statistical learning model can in particular be used in any of the embodiments of the identification system presented above.

[0050] In some embodiments, the statistical learning model is a multi-label classifier.

[0051] In some embodiments, the statistical learning model comprises at least one neural network, preferably of the deep convolutional type.

[0052] In some embodiments, the statistical learning model comprises a hybrid neural network, assembling convolutional and recurrent architectures.

[0053] In some embodiments, the statistical learning model comprises several classifiers trained on different databases. It is thus possible to combine the predictions of these models in order to improve the performance and robustness of the predictions.

[0054] In certain embodiments, the learning method comprises a step of increasing the data of the main database during which certain photographs are synthesized and / or duplicated by applying at least one geometric and / or photometric transformation while retaining the type of each of the defects visible in the photograph. The transformations applied may in particular be rotations, cropping, modifications of the contrast, brightness and / or saturation, to cite only these examples. On the one hand, such an increase in the data makes it possible to multiply the number of photographs and to increase their diversity, which makes it possible to deepen the learning of the statistical learning model, in particular with regard to certain rare defects which are only poorly represented in the initial database.On the other hand, by training the statistical learning model to recognize such defects despite such modifications, the robustness of the statistical learning model is increased.

[0055] In some embodiments, the main database further comprises, for each photograph, information concerning the material of the pipe. Such complementary information provides a complementary parameter to the statistical learning model, making it easier to recognize defects according to the material of the pipe. In particular, certain types of defects appear more frequently in pipes of a certain material.

[0056] In certain embodiments, the main database further comprises, for each photograph, information concerning the location of each defect. This information may in particular comprise the hourly position of the defect, i.e. the angular position of the defect along the circumference of the pipeline. Such complementary information provides a complementary parameter to the statistical learning model, making it easier to recognize the defects depending on their position in the pipeline. In particular, certain types of defects appear more frequently in certain areas of the pipeline.

[0057] The present disclosure also relates to a method for identifying a fault in a pipeline, comprising the following steps: provision of a statistical learning model, configured to determine, on the basis of a photograph of a section of pipeline, the type of each defect recognizable on the photograph from a pre-established list of at least 10 types of defects, route of the pipeline using a robot equipped with a locomotion device and a camera, taking photographs of sections of the pipeline suspected of having one or more defects, and transmission of each photograph to the statistical learning model and determination by the latter of the type of each recognizable defect on each photograph.

[0058] The advantages of this fault identification method arise from the advantages described above for the fault identification system. In addition, this fault identification method may have all or part of the additional characterizations described above regarding the fault identification system.

[0059] In some embodiments, the statistical learning model has been trained using a training method according to any of the preceding embodiments.

[0060] The above-mentioned characteristics and advantages, as well as others, will appear on reading the detailed description which follows, of exemplary embodiments of the identification system and of the proposed learning and identification methods. This detailed description refers to the attached drawings. Brief description of the drawings

[0061] The attached drawings are schematic and are intended primarily to illustrate the principles of the disclosure.

[0062] In these drawings, from one figure to another, identical elements (or parts of elements) are identified by the same reference signs.

[0063] [Fig.l] [Fig.l] is a schematic view of an example of an identification system.

[0064] [Fig.2] [Fig.2] shows this identification system in use on a inspection site. Description of the embodiments

[0065] In order to make the description more concrete, an example of a system for identifying faults in a pipeline as well as the associated training and identification methods are described in detail below, with reference to the attached drawings. It is recalled that the invention is not limited to these examples.

[0066] [Fig.l] represents an example of an identification system 1. This identification system comprises a processing system, taking the form of a central unit 20 housed in a vehicle 3, and a robot 40, which can be stored and transported in the vehicle 3. The vehicle 3 further comprises a control station 5 at which an operator 6 can be installed.

[0067] The central unit 20 comprises a database 21, a statistical learning model 22, a human-machine interface 23 and editing software 24. It also comprises all the usual components useful for its operation, and in particular a processor, a memory and a storage disk.

[0068] The database 21 comprises a very large number, greater than 1,000,000, of photographs of pipeline sections taken from inside the pipeline. In all these photographs, at least one defect is visible and, in at least some photographs, several defects are visible in the same photograph. For each photograph in the database 21, the list of defects visible in this photograph is associated with the photograph in question in the database 21: more precisely, for each defect, the type of defect according to standard NF-13508-2 is specified.

[0069] This standard codifies the various defects that can be encountered in a pipeline and classifies them into categories and subcategories. At the category level, each type of defect is codified using a specific trigram. Here are some examples of types of defects with the associated trigram in parentheses: crack (BAB); surface degradation (BAF); defective connection (BAH); visible soil (BAO); deposits (BBC); adherent deposits (BBB); rupture / collapse (BAC); penetrating connection (BAG); pipe displacement (BAJ); roots (BBA); and connection (BCA).

[0070] The database thus includes at least the trigram of each defect, corresponding to its category within the meaning of standard NF-13508-2, sometimes with the additional indication of its subcategory. In addition, certain neighboring categories can sometimes be grouped into a common class: for example, deposits and adherent deposits can be grouped into a class bearing the label "BBC+BBB".

[0071] In practice, the database 21 is initially constituted from past inspection reports: thus, the type of all the defects recorded in the database 21 has been determined by an operator during a past inspection or, at least, has been verified and validated by an operator.

[0072] In a complementary manner, the database 21 can associate with each defect information concerning the material of the pipeline in which this defect was found as well as information concerning its location within the pipeline, for example its hourly position along the circumference of the pipeline.

[0073] In this example, the database 21 is available locally but it is naturally understood that it could just as easily be hosted in a computing cloud.

[0074] The statistical learning model 22 is a deep convolutional neural network forming a multi-label classifier. As symbolized by the arrow 11, it has been trained on the database 21 so as to be able to determine the type of defect(s) present in a photograph of a section of pipe. In the present example, the statistical learning model 22 has thus been trained to distinguish ten different types of defects and to calculate, from a given photograph, the probability of presence of each of the ten types of defects known by the statistical learning model 22.

[0075] More specifically, in the present example, the multi-label classifier was built from standard architectures by re-training a certain number of layers with images from the database 21.

[0076] Naturally, this training could have been carried out upstream, in a dedicated computer system, separate from the central unit 20, before including the statistical learning model 22 resulting from this training in the processing system 20.

[0077] The statistical learning model 22 is also provided with a thresholding function, with a threshold that can be adjusted in a personalized manner for each type of defect, allowing the statistical learning model 22 to conclude whether or not each type of defect is present by checking whether the probability of the presence of this type of defect is greater than the corresponding threshold.

[0078] In this example, the statistical learning model 22 is available locally but it is naturally understood that it could just as easily be hosted in a computing cloud and remotely interrogated by the processing system 20 via a network connection.

[0079] The human-machine interface 23 comprises a screen and an input device such as a keyboard and / or a mouse. It allows the central unit 20 to transmit information to the operator 6, mainly in the form of images displayed on the screen, and to collect information entered by the operator 6 using the input device. The human-machine interface 23 also comprises a device for controlling the robot 40. It may also comprise other elements such as a loudspeaker and / or a microphone.

[0080] The editing software 24 is software allowing in particular the drafting of inspection reports. It may also include means for printing and / or electronically transmitting documents.

[0081] The robot 40 comprises for its part a locomotion device 41 and an orientable camera 42. The locomotion device 41 here comprises wheels driven by a motor but it could be of a completely different type; it could for example comprise tracks.

[0082] The camera 42 is an RGB (red-green-blue) camera, i.e. a standard camera working in the visible light range; it also has a standard lens, which is not of the fish-eye type and therefore has negligible distortion.

[0083] The robot 40 further comprises all the usual components useful for its operation, and in particular a battery and a communication device. In particular, communication with the central unit 20 can be done wired, using a cable 43, or wirelessly, for example by wifi or radio waves.

[0084] An example of a defect identification method using such a defect identification system 1 will now be described with reference to FIGS. 1 and 2. In particular, [Fig. 2] illustrates the identification system 1 in use at an inspection site 90. The target of this inspection is a wastewater pipe 91 buried under the roadway 92. The pipe 91, having a nominal diameter typically between 200 and 500 mm, extends between a first inspection chamber 93a and a second inspection chamber 93b. This pipe has several defects 94, for example cracks 94a, surface degradation 94b or even a deposit 94c.

[0085] During the inspection, the vehicle 3 hosting the identification system 1 is parked near the first inspection chamber 93a and the robot 40 is introduced into the pipeline 91 through the first inspection chamber 93a. An operator 6 then installs himself at the control station 5 and begins to pilot the robot 40 using the human-machine interface 23: the operator 6 is thus able to control the locomotion device 41 and the camera 42 of the robot 40 and receives the video stream from the camera 42 on a screen. This control-command is represented by the arrow 12 which corresponds to the cable 43.

[0086] The operator 6 then gradually advances the robot 40 into the pipe 91 and monitors the video stream from the camera 42 looking for defects. When he thinks he has spotted a defect 94, the operator 6 can orient the camera in order to better visualize the defect 94: the operator 6 then takes a photograph of the defect 94. When two distinct defects 94 are close to each other, the operator 6 can take a common photograph for the two defects 94.

[0087] As symbolized by arrow 13, this photograph is then transmitted to the statistical learning model 22. The statistical learning model 22 then determines, on the basis of this photograph, the probability of presence of each type of defect known by the statistical learning model 22 on the photograph in question. The statistical learning model 22 then applies its function of thresholding in order to establish a classification proposal for each of the 94 defects recognized by the statistical learning model 22.

[0088] As symbolized by arrow 14, this classification proposal is transmitted to operator 6 via the human-machine interface 23. Operator 6 can then verify this proposal and, depending on the case, confirm or correct it, thus resulting in a definitive classification for each of the defects 94 of the photograph in question.

[0089] As symbolized by the arrow 15, the photograph in question as well as the definitive classification of the defects 94 present on this photograph are transmitted to the editing software 24 in order to progressively write, in a semi-automatic manner, an inspection report, the operator 6 being able to complete and / or correct the latter at any time using the human-machine interface 23.

[0090] Furthermore, as symbolized by the arrow 16, the photograph in question is added to the database 21, in association with the definitive classification of the defects 94 present on this photograph. On this occasion, if the classification proposal established by the statistical learning model 22 has proven to be inaccurate and has therefore been corrected by the operator 6, the central unit 20 records this information in a memory so as to be able to evaluate the statistical learning model and estimate its drift. At regular intervals, for example once every two months, the parameters of the statistical learning model 22 can be adjusted in order to correct this drift.

[0091] The operator 6 can then proceed to search for a new defect 94 in the pipe 91. The steps of the identification method are thus repeated until the robot 40 has finished traveling along the pipe 91. The operator 6 can then finalize the inspection report, recover the robot 40 and store the latter again in the vehicle 3.

[0092] Although the present invention has been described with reference to specific exemplary embodiments, it is obvious that modifications and changes may be made to these examples without departing from the general scope of the invention as defined by the claims. In particular, individual features of the various illustrated / mentioned embodiments may be combined in additional embodiments. Therefore, the description and drawings should be considered in an illustrative rather than restrictive sense.

[0093] It is also obvious that all the characteristics described with reference to a method are transposable, alone or in combination, to a device, and conversely, all the characteristics described with reference to a device are transposable, alone or in combination, to a method. To make the disclosure more concrete, an example of a device is described in detail below, with reference to the attached drawings. It is recalled that the invention is not limited to this example.

Claims

Claims

1. System for identifying defects in a pipeline, comprising a robot (40), provided with a locomotion device (41) and a camera (42), configured to travel along the pipeline (91) and take photographs of portions of the pipeline (91) suspected of having one or more defects (94), a processing system (20), provided with a statistical learning model (22) configured to determine, on the basis of such a photograph, the type of each defect (94) recognized in the photograph from a pre-established list of at least 10 types of defects, and a human-machine interface (23) comprising at least one screen, the processing system (20) being configured to display on the screen the current photograph accompanied by a classification proposal for each of the defects (94) recognizable in the photograph, in which the processing system (20) is configured to receive confirmation or correction of the proposal,entered by an operator (6) via the human-machine interface (23), and to record the final classification for each of the defects (94) in the photograph.,

2. Identification system according to claim 1, wherein the robot (40) is configured to travel through a non-visitable wastewater pipe, having a nominal diameter of less than 1600 mm, preferably between 200 and 500 mm.

3. An identification system according to claim 1 or 2, wherein each photograph comprises at least one defect (94), and wherein at least some photographs comprise several defects (94).

4. Identification system according to any one of claims 1 to 3, wherein the robot (40) is configured to be piloted by an operator (6), wherein the camera (42) of the robot (40) is orientable by the operator (6), and wherein the photographs are triggered manually by the operator (6).

5. An identification system according to any one of claims 1 to 4, wherein the processing system (20) is configured to receive without delay the photographs taken by the robot (40) and transmit them without delay to the statistical learning model (22) for determination of fault types.

6. An identification system according to any one of claims 1 to 5, further comprising a database (21) of photographs comprising at least 1,000, preferably at least 100,000, more preferably at least 1,000,000, photographs of pipeline portions exhibiting at least one defect, each photograph being associated with the type of each of the defects visible in the photograph, and wherein the statistical learning model (22) has been trained on said database.

7. Identification system according to any one of claims 1 to 6, wherein the statistical learning model (22) is a multi-label classifier comprising at least one neural network, preferably of the deep convolutional type.

8. An identification system according to any one of claims 1 to 7, wherein the learning model (22) is configured to determine a probability of presence for each of the types of defects in the pre-established list, wherein the learning model (22) is configured to determine that a particular type of defect is present if it determines that the probability of presence of this type of defect is greater than a predetermined threshold, each type of defect being associated with a personalized threshold.

9. Identification system according to any one of claims 1 to 8, wherein the processing system (20) is configured to further display superimposed on the photograph the position of the recognized defects (94).

10. An identification system according to any one of claims 1 to 9, wherein the processing system (20) is configured to add each new photograph to the database (21), in association with the final classification of defects (94), and wherein the processing system (20) is configured to evaluate the statistical learning model (22) and / or adjust the parameterization of the statistical learning model (22) based on the confirmations or corrections entered by the operator (6).

11. A method of identifying a fault in a pipeline, comprising the following steps: providing a statistical learning model (22), configured to determine, on the basis of a photograph of a portion of pipeline, the type of each defect recognizable on the photograph from a pre-established list of at least 10 types of defects, route of the pipeline (91) using a robot (40) equipped with a locomotion device (41) and a camera (42), taking photographs of portions of the pipeline (91) suspected of having one or more defects (94), transmitting each photograph to the statistical learning model (22) and determining by the latter the type of each defect (94) recognizable on each photograph, display on a screen of a human-machine interface (23) of a classification proposal for each of the defects (94) recognizable in the current photograph, entry by an operator, via the human-machine interface (23), of a confirmation or correction of the proposal, and recording of the final classification for each of the defects (94) in the photograph.