AUTOMATIC METHOD FOR MONITORING AN INSTALLATION COMPRISING A PLURALITY OF SECTIONING DEVICES, PROCESSING TERMINAL AND ASSOCIATED SYSTEM.

Abstract

AUTOMATIC METHOD FOR MONITORING AN INSTALLATION COMPRISING A PLURALITY OF SECTIONING ORGANS, ASSOCIATED PROCESSING TERMINAL AND SYSTEM One aspect of the invention relates to an automatic method (100) for monitoring an installation comprising a plurality of sectioning organs and a plurality of labels each associated with one of said sectioning organs, each sectioning organ being capable of occupying at least two distinct operating states and being referenced by a functional marker, said functional marker being displayed on the associated label, in which the monitoring is carried out from a video stream potentially capturing one or more sectioning organs of the installation, by defining monitoring information associating the functional marker and the operating state of each captured sectioning organ. Figure to be published with the abstract: Figure 2

Description

Title of the invention: AUTOMATIC METHOD FOR MONITORING AN INSTALLATION COMPRISING A PLURALITY OF SECTIONING ORGANS, PROCESSING TERMINAL AND ASSOCIATED SYSTEM. TECHNICAL FIELD OF THE INVENTION

[0001] The technical field of the invention is that of the systematic monitoring of an industrial installation comprising a plurality of materials, organs or equipment capable of occupying at least two distinct functional states.

[0002] The present invention relates to a method for remotely monitoring an operating state of a sectioning member, such as a valve or a switch, of an industrial installation. More particularly, the present invention relates to a monitoring method based on automatic image recognition. The present invention further relates to a processing terminal and a system for implementing the monitoring method.

[0003] The invention finds in particular an application for the monitoring and management of a nuclear installation or any industrial installation implementing a network, or line, of water, gas or electricity, comprising sectioning devices. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0004] Current industrial installations reach a very high degree of complexity due to the number and diversity of functional positions and / or the equipment constituting them.

[0005] In order to ensure the safety of these installations, as well as their proper functioning, regular monitoring is generally carried out by an operator, called a patrolman, carrying out a round following a determined route.

[0006] During this round, the operator ensures the identification of the equipment to be checked by reading and writing, on paper or in an electronic file, a coded identification mark attached to this equipment and referenced in a database relating to the installation. Then, the operator records, checks and possibly modifies the functional state of the equipment.

[0007] The equipment is for example a sectioning member capable of taking two visually distinct operating states, such as a valve operable between an open state and a closed state.

[0008] The results of the readings associated with the identified organs are transmitted, after validation by the operator, to the data centralization station such as the room. management of the installation processes.

[0009] This type of monitoring, when carried out manually, is costly in terms of time and personnel. The number of checks that can be carried out is therefore limited and insufficient to enable a reactive response regarding the configuration of the process or installation.

[0010] To remedy this, it is known to equip each component of the installation with a sensor capable of recording the operating state of the component and transmitting this state to a control center. However, this solution is particularly expensive and time-consuming to implement.

[0011] In order to avoid modifying the installation, document CN111028225A describes the use of a camera oriented so as to acquire the image of a valve of the installation, and of an image analysis module in which automatic image processing algorithms are applied to the acquired image to determine the state of the valve. These algorithms use several neural networks previously trained on learning bases. They are successively implemented by the image analysis module, to determine, in the acquired image, a) a first region of interest corresponding to the valve, b) in this region of interest, a second region of interest corresponding to the handle of the valve, c) in this second region of interest, the angle made by the handle with respect to a closed orientation and the closed (or open) state of this valve, and d) the open or closed state of the valve.

[0012] This solution requires a large number of steps to obtain the operating status of a single valve. It also has the disadvantage of not integrating a prior step of automatic identification of the component to be controlled. By "automatic identification" is meant the determination of the reference - or registration - assigned to this component and which allows its location within the installation.

[0013] However, this step, when it is manual, is subject to a certain number of errors. The operator may, for example, make an erroneous identification of the equipment(s) whose operating status he wishes to check, mainly due to data entry errors. These errors can lead to harmful consequences: malfunction of the installation leading to a partial or total shutdown, endangerment of the operators, etc.

[0014] To improve the reliability of the identification task and ultimately the reliability of monitoring, solutions provide for the installation of paper or digital identification media on each organ to be controlled. The operator responsible for the round is then equipped with a reading system specifically dedicated to reading these identification media.

[0015] Document CN216771532U thus describes a barcode type label in one dimension or a two-dimensional bar code or QR code (according to the acronym “Quick Response code” in English) installed on each part of the installation.

[0016] Such identification labels can, however, deteriorate over time. To remedy this, patent EP0650169B1 describes an electronic identification medium taking the form of an electronic circuit comprising a programmable memory in which the reference of the component in the installation is written.

[0017] This latter solution, although robust, remains however expensive to implement.

[0018] There thus remains a need for a monitoring system and an associated method which is at the same time more reliable, simpler and less expensive to implement and deploy than existing solutions.

[0019] This need exists for monitoring sectioning devices having at least two visually distinct operating states, such as valves, switches, etc. Summary of the invention

[0020] The invention offers a solution to the problems mentioned above, by using a reliable identification medium that is already an integral part of the installation and by making it possible to carry out the identification and recognition of the functional state of the organ to be controlled jointly, by using a single artificial intelligence trained on a training base specific to the organs to be controlled and to the identification labels of the installation.

[0021] A first aspect of the invention relates to an automatic method for monitoring an installation comprising a plurality of sectioning members and a plurality of labels intrinsic to the installation, each label being associated with one of said sectioning members, each sectioning member being capable of occupying at least two distinct operating states and being referenced by a functional reference, said functional reference being displayed on the associated label, said method comprising the following steps: • Reception of a video stream potentially capturing one or more sectioning devices of the installation to be monitored, said video stream having an acquisition frequency restoring real time, and being represented by a time series of images, the method comprising the following steps, carried out upon receipt of each image of the time series of images: • Use, on the image concerned, of an artificial intelligence trained offline on a training database to obtain a first set of characteristics for localization and classification of the state of operation relating to each sectioning organ present in the image concerned, and a second set of location characteristics relating to each label present in the image concerned, • If the first and second sets each have at least one location, • Merging the features of the first and second sets to obtain a set of pairs of locations, each pair of locations associating one of the locations of the first set with the closest location of the second set in the image, and corresponding to the sectioning organ of the location of the first set and its associated label, • For each pair obtained, identification of the associated label from the location of the second set of said pair, and association of the identification obtained with the classification of the operating state corresponding to the sectioning member of said pair, said association defining monitoring information for the sectioning member corresponding to said pair, said monitoring information comprising the operating state and the reference of said sectioning member.

[0022] Thus, the monitoring of the installation is carried out by obtaining monitoring information for each sectioning member captured by the video stream, this monitoring information associating the operating state and the functional reference of said sectioning member.

[0023] Thanks to trained artificial intelligence, we obtain jointly - or simultaneously - the location and operating status of all the sectioning devices present in an image of the video stream, as well as the locations of all the labels present in this image.

[0024] The information thus obtained is then merged and combined with each other to obtain for each sectioning organ present and identifiable in the time series of images a pair {operating state - functional reference], also called monitoring information.

[0025] It should be noted that these merging and combining steps are carried out only for each sectioning organ present and identifiable in a certain manner in the processed image.

[0026] This makes it possible to obtain, in a reduced number of steps, complete monitoring information (including the operating status and registration of the sectioning device).

[0027] The process is thus simpler than existing solutions. One advantage is that it is easily implementable in an on-board processing terminal, which is a key point for its deployment in an installation.

[0028] The method does not require a reading system dedicated to identifying the sectioning member, which implies simplicity and reduced cost of implementation.

[0029] It should be added that the use of a label intrinsic to the installation and containing the functional reference of the sectioning member, compared to a label added specifically to carry out the method, reduces the risk of making an error in identifying the sectioning members. Indeed, the label intrinsic to the installation is placed in the installation at the time of assembly of the sectioning member, and the functional reference that it displays provides an individual (unique) and certain identification of the sectioning member.

[0030] The use of the intrinsic label also avoids the tasks linked to the referencing of the organs to be controlled: creation of a database referencing each organ to be controlled, installation of labels, validation of the label-reference correspondence, etc. An advantage is that the time, and the cost, of deploying the monitoring method are reduced.

[0031] In addition to the characteristics which have just been mentioned in the preceding paragraph, the monitoring method according to the first aspect of the invention may have one or more complementary characteristics among the following, considered individually or according to all technically possible combinations.

[0032] The method comprises the following steps performed after the time series of images has been processed: • Consolidation of monitoring information obtained for each sectioning organ captured in said time series of images, • Transmission of consolidated monitoring information to a central terminal in the installation and / or to a human-machine interface.

[0033] The use of a time series of images to confirm the monitoring information makes it possible to take into account variations in scale, movement, lighting which potentially can induce a determination of the operating state or erroneous identification. The method thus makes it possible to improve the estimation of the monitoring information and is more robust.

[0034] The step of consolidating the monitoring information can then comprise the following sub-steps: • Sorting the monitoring information obtained for all the images in the image time series, to obtain a set of monitoring information associated with each sectioning organ present in the image time series, • determination of characteristics of said set of information of sur- surveillance, • selection of final monitoring information as monitoring information of the sectioning device on the basis of said characteristics.

[0035] Thus, the monitoring information obtained over a time series of successive images is analyzed and the final monitoring information is determined using this analysis. The final monitoring information is thus determined more reliably. This also allows monitoring information to be obtained even if, for reasons of lighting or environment (occlusion, texture, etc.), no classification could be given on a given image of the set of images.

[0036] The step of consolidating the monitoring information may also include the following sub-steps: • Merging the label locations obtained for all images in the image time series, to obtain a set of label locations associated with each pair of locations obtained in the image time series, • Determining characteristics of said set of label locations, • Selecting a final label location as the label location of said pair based on said determined characteristics.

[0037] Thus, the label locations relating to the same label, obtained over a time series of successive images, are analyzed, and on the basis of this analysis, the final label location which will be used to read (identify) the functional marker is determined. This thus makes it possible to take into account variations in scale from one image to another, and to choose the label location which will be the most easily identifiable. The robustness and reliability of the method are thus improved.

[0038] Said characteristics can then comprise the number of occurrences of the same classified operating state, and the final operating state is the operating state corresponding to the highest number of occurrences.

[0039] The number of images in the time series of images can then in practice be between 2 and 30, preferably equal to 10.

[0040] The series of images corresponds to images acquired at a frequency of between 1 image per second and 30 images per second, preferably between 10 images per second and 20 images per second, preferably equal to 10 images per second.

[0041] Artificial intelligence is a neural network trained in a supervised manner on a training basis to provide a location and operating state of each sectioning organ present in an image, and to provide a location of each label present in said image.

[0042] The training base comprises a plurality of images of training sectioning members, each training sectioning member being represented in each of the operating states to be provided, and a plurality of drive label images, each image of the plurality of drive sectioning member images being associated with information relating to the location and operating state of each sectioning member present in the image, and each image of the plurality of drive label images being associated with the location of each drive label present in said image.

[0043] Preferably, the training base is acquired on at least one installation.

[0044] Alternatively, a portion of the training base is acquired on at least one installation, and another portion of the training base is acquired outside the context of the installation.

[0045] The term "outside the context of the installation" means that the images of the training base are not acquired in the installation. It is sufficient to acquire images of the sectioning devices alone, without the context of the equipment located behind and around these devices in the installation. An advantage is that it is simple to construct this training base, and simple to increment it.

[0046] The training base can advantageously be artificially increased from images of sectioning members and / or images of training labels acquired on at least one installation and / or outside the context of an installation.

[0047] The functional marker is encoded with alphanumeric characters, and the step of identifying the label is carried out using an automatic alphanumeric character recognition algorithm.

[0048] When the functional marker is coded with alphanumeric characters, the identification step comprises, after the use of the automatic character recognition algorithm, a step of verifying the result of the recognition algorithm with respect to predetermined grammar rules to confirm the functional marker.

[0049] This therefore makes it possible to confirm that the code read is indeed a functional reference and not a code relating to other equipment / functions.

[0050] Thus, the functional marker is read simply regardless of the orientation of the label, without recourse to a training base specific to the installation.

[0051] The method comprises, after the step of associating the functional reference and the operating state, a step of transmitting said association to a central terminal and / or to a display-writing means.

[0052] Thus, the results are transmitted to the central terminal of the installation for updating the functional plan and / or the general file of the installation, and / or to a patrolman who can validate, or follow, the result given by the process.

[0053] The sectioning member is a quarter-turn valve operable between an open state and a closed state.

[0054] The database can then contain less than 20,000 labels.

[0055] A second aspect of the invention relates to a processing terminal for monitoring an installation, configured to execute the method according to the first aspect of the invention.

[0056] The processing terminal may be portable.

[0057] A third aspect of the invention relates to a system for monitoring an installation comprising a plurality of sectioning members arranged in determined locations of the installation, each sectioning member being capable of occupying at least two distinct operating states and being referenced by a functional reference, and a plurality of labels, each label coding the functional reference of one of the sectioning members and being arranged in a determined location of the installation adjacent to a determined location for said sectioning member, comprising: • a processing terminal according to the second aspect of the invention, • a module configured to acquire a stream of images from the installation, • a lighting system.

[0058] The system may comprise a display-write module communicating with the processing terminal and made available to an operator, the video module, the processing terminal and said display-write module being adapted to be worn by said operator.

[0059] The system may comprise a central terminal comprising means for transmitting and receiving data wirelessly to the processing terminal.

[0060] The invention and its various applications will be better understood upon reading the following description and examining the accompanying figures. BRIEF DESCRIPTION OF THE FIGURES

[0061] The figures are presented for information purposes only and in no way limit the invention. • [Fig.l] is a block diagram illustrating the sequence of steps of a method for monitoring an installation according to the invention, • [Fig.2] is a schematic representation of captured sectioning organs subjected to the monitoring process. • [Fig.3] shows a schematic representation of an example of a sectioning member to be monitored, said sectioning member being in an open operating state, • [Fig.4] shows an example of a label associated with the sectioning device of [Fig.3], • [Fig.5] shows a schematic representation of an image obtained at the end of the second stage of the process of [Fig.l], • [Fig.6] shows a schematic representation of an image obtained from the image of [Fig.5] at the end of the third step of the method of [Fig.l], • [Fig.7] is a block diagram illustrating the sequence of sub-steps of the fourth step of the process of [Fig.l], • Figures 8A and 8B are block diagrams illustrating the sequence of sub-steps of the sixth step of the method of [Fig.l], • [Fig.9] shows a schematic representation of a processing terminal configured to perform the method of [Fig.l], and of a system for monitoring an installation DETAILED DESCRIPTION

[0062] The invention relates in particular to a method for monitoring an installation comprising a plurality of sectioning members and labels associated with said sectioning members, each sectioning member being capable of taking at least two distinct operating states and being referenced by a functional marker.

[0063] By “distinct operating states” is meant optically distinguishable, or, in other words, visually distinguishable.

[0064] By “functional reference” is meant a unique coded reference allowing the location on the installation of the sectioning device to which it is assigned and its functional description.

[0065] The monitoring of the installation is carried out when monitoring information is determined for each sectioning member, this monitoring information being defined as the association of the functional reference and the operating state of this sectioning member.

[0066] [Fig.l] is a block diagram illustrating the sequence of steps of an automatic method 100 for monitoring an installation according to the invention.

[0067] The monitoring is carried out from an FV video stream potentially capturing one or more sectioning devices of the installation.

[0068] For this, the method 100 begins with a step S101 of receiving the video stream FV.

[0069] This FV video stream has an acquisition frequency restoring real time, and is represented by a time series of images Ib Ij, In.

[0070] The acquisition frequency is preferably between 1 image per second and 30 images per second, preferably between 10 images per second and 30 images per second, and preferably equal to 15 images per second.

[0071] This FV video stream can be acquired directly by a video module (not shown in [Fig.l]) comprising a camera, an acquisition card and means of transmission to a processing terminal.

[0072] This video module can be manipulated (oriented) by a patrolman carrying out surveillance along a determined route in the installation, this route making it possible to capture in several video streams all of the sectioning devices of the installation. In this case, the video module can be installed on an arm held by the patrolman, or it can be installed on an accessory worn by the patrolman, such as glasses.

[0073] This video module can alternatively be installed on a mobile robot traveling through the installation and programmed to capture the installation's sectioning devices in the video stream.

[0074] Alternatively, the reception S101 of the video stream FV first comprises a reception of a primary video stream (not shown in [Fig.l]) acquired by the video module then a conversion of this primary video stream to obtain the video stream FV used by the method 100.

[0075] The primary video stream then has an acquisition frequency higher than the acquisition frequency of the FV video stream used by the method 100. The frequency of the primary video stream is thus, for example, between 30 images per second and 60 images per second.

[0076] The time series of images L, Ij, L representing the video stream FV preferably comprises, when the frequency of the primary video stream is 30 images per second, between 1 and 30 images, preferably 10 images.

[0077] [Fig.2] is a schematic representation of sectioning organs O, HO;, O; +i captured in the video stream FV and subjected to the monitoring method 100.

[0078] By “captured in the FV video stream”, we mean that these sectioning members are each present in at least one image Ij of the FV video stream. They can be present individually (a single sectioning member in an image) or together (several sectioning members in an image).

[0079] The sectioning devices O, h O;, Oi+i were installed in specific locations of the INST installation, along the LIGN line circuits of this INST installation.

[0080] The sectioning organs O, h O;, Oi+i can be valves, switches, or other.

[0081] Whatever the nature of the sectioning members, the shape, size, and / or color of each sectioning member Ou, Oi, Oi+i may vary from one sectioning member to another.

[0082] Thus, if the sectioning members are valves, these can be of different types to serve different functions: isolation valves, stop valves, etc. In other words, the method according to the invention is compatible with different types of valves, as long as these types have at least two visually distinct operating states.

[0083] As mentioned previously, each sectioning member Oi-1, Oi, Oi+i has an operating state Fm, F;, Fi+i selected from at least two distinct operating states. In this same [Fig.2], the functional states Fm, F;, Fi+i of the sectioning members Oi-1, Oi, Oi+i are two in number, and are represented schematically by the graduation 0 (members O and O;) and 1 (member O i+i)-

[0084] [Fig. 3] shows a schematic representation of an example of a valve. The valve has a handle P that can be operated between a first position parallel to the X axis of the valve (shown in [Fig. 3]) and a second position oriented at 90° relative to the first position (not shown in [Fig. 3]). These two positions correspond to the two distinct functional states represented respectively by the graduations 1 (open state) and 0 (closed state) in [Fig. 3].

[0085] As mentioned previously, each sectioning member is furthermore associated with, or referenced by, a functional reference denoted RE i, RF;, RFi+i (respectively for the sectioning members O, h Oi, Oi+i) which is unique (or individual) and allows the location and functional description of said sectioning member Om , Oi, Oi+1.

[0086] This functional reference RE i, RF;, RFi+i is generally recorded in a central file stored in a central terminal (or central monitoring terminal, not shown in [Fig.2]) of the INST installation with other information relating to the sectioning device (for example, the supplier, the date of the last inspection, etc.).

[0087] The functional reference is intrinsic to the installation, that is to say that it is an integral part of the installation. It is implemented from the design of the installation to describe each sectioning device. Thus, to implement the method 100, there is no need to carry out a prior task of assigning an identifier to the sectioning devices. An advantage is that the deployment of the monitoring method 100 is faster and less expensive compared to solutions where it is a question of adding labels or any device carrying the identification function dedicated to automatic monitoring.

[0088] In [Fig.2], the functional reference RFn, RF;, RFi+i is preferably coded with alphanumeric characters.

[0089] [Fig.2] schematically represents the labels Em, E, Ei+i associated with the sectioning members O, h Oi, Oi+i.

[0090] The labels Ei i, E, Ei+i are said to be “intrinsic to the installation” because they are placed in the INST installation at the time of assembly of the sectioning devices. to which they are associated and carry the functional reference assigned to these sectioning organs.

[0091] Each label E; associated with a sectioning member O; contains functional reference RF; assigned to this sectioning member O; and is affixed visibly and as close as possible to this sectioning member O;. The term “visibly” means in such a way that a patrolman can read the label E;. The term “as close as possible” designates a location such that the patrolman can unambiguously connect the label E; to the sectioning member O; with which it is associated.

[0092] [Fig.4] shows an example of a label E;. This is for example in the form of a piece of black plastic with the functional reference RF; written in white characters. Such a label is glued, or screwed, as close as possible to the sectioning member O; with which it is associated.

[0093] Naturally, several types of labels can be used. For example, labels with a white background and black characters and labels such as those shown in [Fig.4] can be used together.

[0094] The block of steps B1 consisting of steps S102, S103, S104, S105 and S106 is carried out as an extension of step S101 of receiving the video stream FV, on each image Ij, j= 1 to n, of the time series of images L, Ij, L of this video stream FV.

[0095] The block of steps B1 aims to provide, for each image Ij of the time series of images, a set Ens4j of monitoring information ISO;, each monitoring information ISO; of the set Ens4i being relative to each sectioning member Oi detected in the image Ij. The monitoring information ISO, H ISO;, SOi+i are illustrated in [Fig.2].

[0096] It should be noted that the block of steps B1 is carried out in a duration less than the time separating two successive images Ij, Ij+1 of this time series of images. Thus, this block of steps B1 is carried out in real time. In other words, the monitoring information relating to each image is considered to be delivered in real time.

[0097] Within the block of steps B1, and with reference to [Fig.2], step S102 consists of using, on an image Ij of the image time series, an artificial intelligence previously trained offline on a training basis, to simultaneously obtain characteristics relating to the sectioning members O, HO;, Oi+i and to the labels present in the image.

[0098] Thus, a single trained artificial intelligence can simultaneously detect several sectioning organs in the image as well as the elements (labels) allowing them to be identified. Detection is thus rapid.

[0099] [Fig.5] shows a schematic representation of an image Ij obtained at the end of step S102 and in which a sectioning member Oi and two labels E;, Ei+i were captured.

[0100] The characteristics relating to the sectioning members present in the image Ij form a first set Enslj of characteristics, and the characteristics relating to the labels E;, Ei+i present in the image Ij form a second set Ens2j of characteristics.

[0101] The characteristics of the first set Enslj comprise location characteristics SO; of each sectioning member present in the image I,, and classification characteristics FO; of each sectioning member Oi present in the image Ij.

[0102] By location characteristic SO;, we mean for example a bounding box surrounding the region of the image corresponding to the sectioning member O;.

[0103] By classification characteristic FO;, we mean the classification of the operating state F; of this sectioning member O;. When the sectioning member O i is a valve, the classification FO; of the operating state E is either an open state 1, or a closed state 0, or an indeterminate state NA distinct from the open state 1 and the closed state 0. In [Fig.5], the classified operating state FO; is the open state 1.

[0104] It should be noted that in this first set Enslj of characteristics, each classification characteristic FO; corresponds to a location characteristic SO i, (and vice versa), the two characteristics then being linked to the same sectioning organ O;.

[0105] The characteristics of the second set Ens2j comprise localization characteristics SE;, SEi+i of the labels present in the image Ij. These localization characteristics can be bounding boxes surrounding each region of the image corresponding to a label E;, Ei+i.

[0106] The artificial intelligence is preferably an artificial neural network trained in a supervised manner on the training basis to provide, from an image Ij, a location SO; of each sectioning member O; present in the image Ij and a location SE; of each label E; present in the image Ij, and for said location SO; of the sectioning member O;, a classification FO; of the operating state of the corresponding sectioning member O;.

[0107] In step S102, a convolutional neural network or CNN (acronym for "Convolutional Neural Network" in English) can be used for the neural network.

[0108] The training base preferably comprises a plurality of images of drive sectioning members, each drive sectioning member being represented in each of the operating states to be provided, and a plurality of images of drive labels. The drive sectioning members com preferably take several types of drive members (several types of drive valves, or several types of drive members).

[0109] The training sectioning members advantageously have morphological (shape, color) and arrangement (orientation, lighting) differences between them. The images of the training base are advantageously acquired with a variety of lighting conditions, distance to the training sectioning member, etc. Thus, the performance of the trained neural network is more robust with respect to the type of sectioning member and / or the shooting conditions.

[0110] Advantageously, the images of the training base are acquired in the context of at least one installation.

[0111] “In the context of an installation” means that the context of the installation is present on the images of the training base, in the environment of the training sectioning members and the training labels. The images of the training base are then acquired on the installation.

[0112] Preferably, the images of the training base are acquired on three separate installations.

[0113] Tests carried out within the framework of the invention have shown that the variety of learning is then sufficient to generalize artificial intelligence well.

[0114] Advantageously, a part of the images of the training base is acquired in the context of one or more installations and another part of the images of the training base is acquired outside the context of an installation.

[0115] By "outside the context of an installation" is meant that the images of the training organs or training labels do not include the other elements of the installation (lineage circuits or other element). Thus, the training base can be enriched to provide missing diversity, in a non-invasive manner, that is to say without needing to travel through the installation or interfere with the operation and management of the installation. Furthermore, the training base makes it possible to obtain a better generalization of the performance of the neural network compared to a training base acquired in the sole context of the installation. The monitoring method is thus easily adaptable to changes in installation.

[0116] The training base can furthermore be advantageously artificially increased from images of sectioning organs and / or images of training labels acquired on at least one installation and / or outside the context of an installation. This makes it possible to train the neural network simply and quickly on a greater variety of situations.

[0117] Thus, the adaptation of the method 100 to a new installation is quick and inexpensive to carry out.

[0118] Each image of the plurality of drive sectioning member images is associated with information relating to the location and operating state of each sectioning member present in the image.

[0119] Each image of the plurality of training label images is associated with information relating to the location of each training label present in said image.

[0120] The information relating to the location and operating status of the drive sectioning members and to the location of the drive labels are called instantiations or labelings. They are obtained for example manually, by annotations of the images of the drive base.

[0121] Advantageously, the training base includes less than 20,000 labels. This number is sufficiently low so that the training base is quick to build.

[0122] The training base may alternatively comprise a single plurality of images including the training sectioning members and the training labels.

[0123] The training base may alternatively comprise a first plurality of images of training sectioning members, a plurality of images of training labels, and a plurality of images including training sectioning members and training labels.

[0124] With reference to [Fig.2], the monitoring method 100 continues with a step S103 which consists of merging, i.e. pooling, the characteristics of the first and second sets Ens2j and Ens3j to detect whether at least one potentially identifiable sectioning member O; is actually present in the image Ij.

[0125] To do this, we can determine the number NsOj of location characteristics relating to the sectioning members, and the number Nsej of location characteristics relating to the labels.

[0126] If the number NsOj of location characteristics relating to the sectioning members and the number Nsej of location characteristics relating to the labels are both null, the response is considered to be negative Dneg, i.e. no sectioning member is present in the image Ij, or if it is present, it is potentially missing information (the associated label) likely to identify it. No other step is then carried out on the image Ij. The method then continues with step S102 which is repeated on the image Ij+i following the image Ij, and so on on the following images of the time series of images, up to the last image L of this series.

[0127] Otherwise, that is, if this number NsOj of loca characteristics If the number of location features relating to the sectioning members and the number Nsej of location features relating to the labels are both unaffected, then the response is considered positive Dpos, i.e. one or more potentially identifiable sectioning members are detected in the image. The method then continues with step S104.

[0128] Step S104 consists of merging the location characteristics SO;, SE; of the first and second set Enslj, Ens2j to obtain a set Ens3j of location pairs.

[0129] [Fig.6] shows a schematic representation of the image Ij obtained at the end of step S104, in which a set Ens3j of location pairs has been obtained.

[0130] In the set Ens3j, each pair PO; of locations associates one of the locations of the sectioning organ SO; with the closest label location SE;. In [Fig.3], the set Ens3j only comprises a single pair POi.

[0131] For this, a calculation of the distance between the bounding box of the sectioning member and each bounding box of labels can be carried out, the smallest distance then being used to determine the closest bounding box of labels.

[0132] Naturally, other methods may be used in combination or not with this distance calculation and distance comparison. For example, in addition to the distance comparison, it is possible to perform a comparison of the size of the bounding boxes in the image, the largest size then being chosen to determine the closest bounding box. It should be noted that, in this case, the labels are all the same size and are positioned either strictly vertically or strictly horizontally throughout the installation.

[0133] The method 100 continues with a step S105, illustrated in [Fig.l], consisting of determining, for each location pair PO; obtained in the set Ens3j, the monitoring information ISO; relating to the sectioning member O; corresponding to said pair PO;.

[0134] For this, step S105 comprises sub-steps S1051, S1052 and S1053, illustrated in [Fig.7],

[0135] The first sub-step S1051 consists of identifying the functional marker RF; of the sectioning member corresponding to said POi pair from the label location SE; of said pair. When the functional marker is coded with alphanumeric characters, the identification advantageously consists of the use of a character recognition algorithm or OCR (acronym for “Optical Character Recognition” in English). The OCR algorithm preferably uses a trained recurrent neural network or RNN (acronym for “Recurrent Neural Network” in English). Such trained algorithms are available free of charge, and do not require the construction of a training base specific to the INST installation. The identification of the sectioning organ is thus simple, quick and easy to implement.

[0136] The result of the OCR character recognition algorithm can be checked with predetermined rules, called grammar rules, to confirm that it is a functional reference. This avoids using an erroneous code as a functional reference because it relates to other functions / equipment. This improves the reliability of automatic identification.

[0137] Sub-step S1051 is extended by a step S1052 consisting of selecting the classification FO; of the operating state corresponding to the location SO; of the sectioning member of said pair.

[0138] Sub-step S1052 is then extended by a sub-step S1053 consisting of associating the RF identification; of the label location, i.e. the functional marker corresponding to the sectioning member of said pair, with the selected FO operating state.

[0139] Thus, at the end of step S105, a set Ens4j of ISO monitoring information; is determined, each piece of ISO monitoring information; corresponding to the association of the functional reference and the operating state of each sectioning member detected in the image h.

[0140] After step S105, the method may comprise a step S106 of transmitting the set Ens4j of ISO monitoring information; determined in the image Ij to a processing terminal and / or to a human-machine interface.

[0141] The human-machine interface can be implemented on a display means made available to the patrolman.

[0142] Once transmitted to the processing terminal, the monitoring information can be used to update a functional plan of the installation (which is for example displayed on a display screen) and / or update the general file of the installation. This file is for example part of installation management software. During this update, an operating anomaly and a malfunction alert can be carried out by the central terminal.

[0143] When the monitoring information is transmitted to a patrolman, the latter can confirm it, thus carrying out a double validation. The monitoring method 100 thus provides help, assistance, to the patrolman.

[0144] The method 100 then continues with step S102 which is repeated on the image Ij+1 following the image Ij, this up to the last image L of the time series of images.

[0145] Once all the images of the time series L, Ij, L have been processed, a set Ens4FV of monitoring information relating to the time series of images is obtained, which comprises the plurality of sets Ens4j of monitoring information obtained for each image Ij where a sectioning organ has been detected.

[0146] The method 100 can then be extended by a step S107 of consolidation (or confirmation) of the monitoring information obtained in the set Ens4FV of monitoring information relating to the time series of images.

[0147] With reference to [Fig.8A] and [Fig.8B], this step S107 may comprise sub-step S1071 and / or sub-step S1072. The order of sub-steps S1071 and S1072 is irrelevant.

[0148] Substep S1071 aims to consolidate / confirm the operating state information in each ISO monitoring information; of the set Ens4FV.

[0149] For this, the sub-step S1071 comprises a sub-step S1071A, illustrated in [Fig.8A], consisting of sorting the monitoring information obtained in the set Ens4FV of monitoring information relating to the time series of images, to obtain sets Ens5; of monitoring information, each set Ens5; being associated with one of the sectioning members O; present in the time series of images.

[0150] Still with reference to [Fig.8A], sub-step S1071A is extended by a sub-step S1071B consisting of determining characteristics for each set Ens5i of monitoring information.

[0151] These characteristics may include the number of occurrences of the same monitoring information.

[0152] Alternatively, these characteristics may comprise the presence of the same monitoring information, i.e. the same pair {operating state-functional reference] on several consecutive images (i.e. on several consecutive positions in the set Ens5;), for example on five consecutive images.

[0153] Sub-step S1072B is extended by a sub-step S1072C, also illustrated in [Fig.8A], consisting of selecting, for each set Ens5i, a final monitoring information as monitoring information of the sectioning member on the basis of the characteristics determined in step S1072B.

[0154] When the characteristics include the number of occurrences of the same monitoring information, the final monitoring information selected is the monitoring information which has the greatest number of occurrences.

[0155] When the characteristics include the presence of surveillance information on five consecutive images in the set Ens5i, the final surveillance information selected is the surveillance information which is present on these five consecutive images.

[0156] Sub-step S1072 aims to consolidate / confirm the information of the functional reference in each ISO monitoring information; of the set Ens4FV.

[0157] For this, sub-step S1072 comprises a first sub-step S1072A, illustrated in [Fig.8B], consisting of merging the label locations obtained for the set Ens4FV of surveillance information relating to the time series of images, to obtain sets Ens6; of label locations, each set Ens6; being associated with a label present in the time series of images.

[0158] Substep S1072B is performed for each set Ens6; of label locations. It consists of determining characteristics of said set Ens6; of label locations.

[0159] One of the characteristics may be the area (in square pixels) covered by the bounding box surrounding the label.

[0160] Substep S1072C which follows substep S1072B is also carried out for each set Ens6i of label locations. It consists of selecting, on the basis of the determined characteristics, a final label location as the label location of the label associated with said set Ens6i.

[0161] For example, when the extracted feature is the area of ​​the bounding box, the final label location is the one with the largest area.

[0162] Thus, the chances of correctly identifying said label are improved and variations in scale or lighting between the images in the time series of images are robustly taken into account.

[0163] At the end of step S107, a consolidated set of monitoring information Ens4CFV is obtained, in which each piece of monitoring information relates to each sectioning member present in the time series of images, and has been determined with increased reliability compared to the set Ens4FV.

[0164] Step S107 can then be followed by a transmission step S108 to the central terminal and / or the human-machine interface of the consolidated monitoring information Ens4CFV for each sectioning device present in the time series of images.

[0165] The human-machine interface can be implemented on a display-writing means, such as a tablet, made available to the patrolman.

[0166] The transmission step S108 can replace the step S106. In this case, only the consolidated monitoring information is transmitted to the central terminal.

[0167] The method 100 can then comprise a step S109 illustrated in [Fig.l], of validation by the patrolman, using the display-write means, of the consolidated monitoring information Ens4CFV.

[0168] During this validation step S109, the patrol officer can confirm the monitoring information obtained automatically by steps S101 to S108 of the method, or he can invalidate it.

[0169] In the event that the patrolman does not confirm the surveillance information obtained automatically for one of the sectioning organs Oi present in the image series, the validation step S109 may comprise an additional sub-step consisting of carrying out the following actions: • Acquire a first image centered on the sectioning organ for which the monitoring information has not been confirmed by the patrolman, • Acquire a second image centered on the label associated with said sectioning organ, • Use artificial intelligence on the first and second images to obtain respectively location and classification characteristics relating to said sectioning organ, and location characteristics of the associated label, • Identify the functional reference of the associated label from the location of said label, • Combine the classification of the operating state and the functional reference to obtain the monitoring information for said sectioning device.

[0170] When the patrolman has manually confirmed the monitoring information for the sectioning devices present in the time series of images, said images can be annotated and saved in a storage space to be used later as images of the artificial intelligence training base. Thus, it is possible to enrich the training base and re-train the artificial intelligence based on feedback. The monitoring method is thus easily adaptable to types of sectioning devices or labels that are rare and potentially difficult to classify.

[0171] Advantageously, the sets of monitoring information Ens4j, Ens4FV, Ens4CFV are obtained using a single neural network trained on a training base specific to the sectioning organs and the labels of the installation.

[0172] The method 100 is thus simple and inexpensive to implement.

[0173] [Fig.9] shows a schematic representation of a processing terminal 11 according to a second aspect of the invention and of a monitoring system 1 according to a third aspect of the invention.

[0174] The processing terminal 11 is configured to execute the method 100.

[0175] This processing terminal 11 comprises a processor 111, a memory (not re shown in [Fig.9]), and a means 112 for transmitting-receiving data to a video module 12.

[0176] It should be noted that the computing power of the processing terminal 11 and the artificial intelligence used in step S102 determine the number of images in the time series of images. At a minimum, this computing power and the artificial intelligence are chosen to process the number of images mentioned above.

[0177] The processing terminal may further comprise a means 113, N of transmission-reception, for example wireless, towards a central terminal 13 of the installation.

[0178] It may comprise a means 114 of transmission-reception towards a display-writing device 14 made available to a patrolman R and in which a human-machine interface is implemented.

[0179] The processing terminal 11 is advantageously portable.

[0180] The processing terminal can advantageously be coupled to an energy storage means 115, such as a portable battery.

[0181] Still with reference to [Fig.9], the system 1 is carried by a patrolman and comprises: • The processing terminal 11 according to the second aspect of the invention, • the video module 12 configured to acquire the FV video stream, • a lighting system 15.

[0182] The lighting system 15 comprises a light source, for example a light-emitting diode, coupled to a means for controlling the processing terminal 11. The control means is configured to activate the emission of light from the light source 15 during the acquisition of the FV video stream by the video module. The light source is preferably installed on the video module so as to enable the lighting of the scene to be filmed. The lighting system 15 thus makes it possible to monitor sectioning members even when they are in a dark environment. It also makes it possible to make the method more robust to variations in lighting conditions.

[0183] The video module 12 is preferably configured to capture a video stream with a field of view including at least one sectioning member and its associated label.

[0184] The video module 12 may comprise a viewfinder that allows the patrolman R to determine how he films the installation. The video module may then be configured to display a viewfinder window in the viewfinder. This viewfinder window allows the patrolman to center the acquisition of the video stream preferentially on a sectioning member and / or on its associated label. The images of the time series of images then comprise a lower density of sectioning members to be controlled and / or fewer context elements, which makes it possible to improve the performance (classification, calculation time) of the artificial intelligence.

Claims

Claims

1. Automatic method (100) for monitoring an installation (INST) comprising a plurality of sectioning members (O;) and a plurality of labels (Ei) intrinsic to the installation, each label being associated with one of said sectioning members (Oi), each sectioning member (Oi) being capable of occupying at least two distinct operating states (0,1) and being referenced by a functional reference (RF;), said functional reference (RF;) being displayed on the associated label (Ei), said method (100) comprising the following steps: - Reception (S 102) of a video stream (FV) potentially capturing one or more sectioning devices (O;) of the installation to be monitored, said video stream (FV) having an acquisition frequency restoring real time, and being represented by a time series of images (Ib Ij, In), the method (100) being characterized in that it comprises the following steps, carried out upon receipt of each image (Ij) of the time series of images (Ib Ij, In): - Use (S102), on the image (Ij) concerned, of an artificial intelligence trained offline on a training database to obtain a first set (Enslj) of location (SE;) and classification (FO;) characteristics of the operating state relating to each sectioning member (O;) present in the image concerned, and a second set (Ens2j) of location (SE;) characteristics relating to each label (Ei) present in the image (Ij) concerned, - If the first and second sets (Enslj, Ens2j) each comprise at least one location (S 103), • Merging (S 104) the characteristics of the first and second sets (Enslj, Ens2j) to obtain a set (Ens3j) of pairs of locations (PO;), each pair (PO;) of locations associating one of the locations (SO;) of the first set with the closest location (SE) of the second set in the image (Ij), and corresponding to the sectioning member (Oi) of the location (SOi) of the first set and to its associated label (Ei), For each pair (POi) obtained, identification (S 105, S1051) of the associated label from the location (SE) of the second set of said pair, and association (S 105, S1052, S1053) of the obtained identification (RF;) with the classification (FO;) of the operating state corresponding to the sectioning member of said pair, said association defining monitoring information (ISO;) of the sectioning member corresponding to said pair, said monitoring information (ISO;) comprising the operating state (FO;, F;) and the reference (RF;) of said sectioning member (O;).

2.

3. Automatic method (100) for monitoring an installation according to the re Claim 1, wherein the method (100) comprises the following steps performed after the time series of images (Ii-In) has been processed: Consolidation (S 107) of the monitoring information obtained for each sectioning organ captured in said time series of images, Transmission (S 108) of the consolidated monitoring information to a central terminal of the installation and / or to a human-machine interface. Automatic method for monitoring an installation according to claim 2, in which the step of consolidating (S 107) the monitoring information comprises the following sub-steps: sorting (S 1071, S1071 A) of the monitoring information obtained for all the images of the time series of images (h -In) to obtain a set of monitoring information (Ens5i) associated with each sectioning member (O;) present in the time series of images, determining (S 1071, S107 IB) characteristics of said set of monitoring information (Ens5i), selection (S 1071, S1071C) of a final monitoring information as monitoring information (ISO;) of the sectioning member (Oi) on the basis of said characteristics.

4. An automatic monitoring method (100) according to one of claims 2 to 3, wherein the step of consolidating (S107) the monitoring information comprises the following substeps: - Merging (S 1072, S1072A) the label locations obtained for all the images in the time series of images, to obtain a set of label locations associated with each pair of locations obtained in the time series of images, - Determining (S 1072, S1072B) characteristics of said set of label locations, - Selecting (S1072, S1072C) a final label location as the label location of said pair on the basis of said determined characteristics.

5. Automatic monitoring method (100) according to one of claims 1 to 4, in which the artificial intelligence is a neural network trained in a supervised manner on a training basis to provide a location (SO;) and an operating state (FO;) of each sectioning member (Oi) present in an image, and to provide a location (SE;) of each label present in said image, the training base comprising a plurality of images of drive sectioning members, each drive sectioning member being represented in each of the operating states to be provided, and a plurality of images of training labels, each image of the plurality of images of drive sectioning members being associated with information relating to the location and operating state of each sectioning member present in the image, and each image of the plurality of images of training labels being associated with the location of each training label present in said image.;

6. Automatic monitoring method (100) according to claim 5, in which the training base is acquired on at least one installation.

7. Automatic monitoring method (100) according to claim 5, in which a part of the training base is acquired on at least one installation, and another part of the training base is acquired outside the context of the installation.

8. An automatic monitoring method (100) according to claim 1 to 7, wherein the functional marker is encoded with alphanumeric characters, and the step of identifying the label is performed using an automatic alphanumeric character recognition algorithm.

9. Automatic monitoring method (100) according to claim 1 to 8, in which the method comprises, after the step of associating the functional marker and the operating state, a step of transmitting said association to a central terminal and / or to a display-writing means.

10. Automatic monitoring method (100) according to claim 1 to 9, in which the sectioning member is a quarter-turn valve operable between an open state and a closed state.

11. Automatic monitoring method (100) according to claim 1 to 10, wherein the number of images in the time series of images is between 2 and 30, preferably equal to 10.

12. Processing terminal (11) for monitoring an installation (INST), configured to execute the method (100) according to one of claims 1 to 11.

13. A processing terminal (11) according to claim 12, wherein the processing terminal (11) is portable.

14. System (1) for monitoring an installation comprising: - a processing terminal (11) according to one of claims 12 to 13, - a video module (12) configured to acquire a stream of images of the installation, - a lighting system (15).

15. System (1) according to claim 14, in which the system (1) comprises a display-write module (14) communicating with the processing terminal (11) and made available to an operator (R), the video module (12), lighting system (15), the processing terminal (11) and said display-writing module (14) being adapted to be worn by said operator (R).

16. System (1) according to one of claims 14 to 15, in which the system (1) comprises means for transmitting and receiving data wirelessly to a central processing terminal (13).

Images