Method for detecting anomaly(ies) in a structure

Abstract

Method for detecting anomaly(ies) in a structure The invention relates to a method (60) for detecting anomaly(ies) in a structure, comprising: a learning phase (62) including the following steps: - obtaining (64) healthy data respectively associated with N distinct sets of conditions of use of said structure; - projecting (66) the healthy data into a latent space of reduced dimension compared to the dimension of said healthy data; - determining (68) the contour of said set of healthy data projected into said latent space; and an operational phase (70) including the following steps: - obtaining (74), by prior measurement (72), test data representative of the current state of said structure; - projecting (76) said test data into said latent space; - detecting (78) at least one current anomaly of said structure as soon as an element of test data is outside said contour.Figure for the abbreviation: Figure 3.

Description

Title of the invention: Method for detecting anomaly(ies) in a structure

[0001] The present invention relates to a method for detecting anomaly(ies) in a structure, said structure carrying at least one sensor for measuring at least one characteristic of said structure, the method comprising at least one learning phase and at least one operational phase.

[0002] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement such a method of detecting anomaly(ies) in a structure.

[0003] The invention also relates to an electronic device for detecting anomaly(ies) in a structure, said structure carrying at least one sensor configured to generate and receive guided waves, the device comprising a learning unit and a test unit.

[0004] The invention also relates to a system for detecting anomaly(ies) in a structure.

[0005] The present invention relates to the field of structural health monitoring (SHM), aimed at detecting and characterizing damage (i.e., anomalies) in structures / infrastructures. Such structural anomalies correspond to changes in the physical and / or geometric properties of the structure in question that are likely to affect its performance and / or reliability.

[0006] Such health monitoring of structures is implemented using integrated sensors, for example piezoelectric sensors designed to emit and receive ultrasonic guided elastic waves.

[0007] The permanent integration of sensors on or within structures / infrastructures to monitor their condition makes it possible in particular to guarantee the safety of load-bearing structures corresponding to thin and / or long mechanical components, such as wind turbine blades, aircraft engine fuselages or components, metal or composite pipes, tension cables, bridge suspension cables, train rails, etc.

[0008] In particular, guided elastic ultrasonic waves, emitted by such sensors, make it possible to detect structural defects leading to discontinuities or variations in geometry, such as cracks, delaminations in composite fuselages, corrosion leading in particular to a loss of thickness in metals, etc., at an early stage and thus to monitor them for decades.

[0009] Such detection is generally well controlled in the laboratory where external effects are limited, controlled and calibrated.

[0010] The main challenge associated with such a structural diagnosis, in real conditions, is linked on the one hand to the presence of epistemic uncertainties on the structure or the instrumentation such as the positions of the sensors, the properties of the sensors, the elastic or geometric properties of the structure, and on the other hand to the presence of unsupervised (i.e. unknown) external influencing effects evolving with distinct temporal dynamics.

[0011] It should be noted that "epistemic uncertainties" refers to uncertainties due to a lack of knowledge that can be reduced by attempting to acquire more knowledge through, for example, data collection, expert assistance, accelerated testing, etc. It should be noted that in the context of SHM, reducing these uncertainties is generally prohibitively expensive, hence the use of the term uncertainties.

[0012] Such unsupervised external influencing effects can be classified on the one hand into so-called "fast" effects, exhibiting a variation on a time scale ranging from the hour to the day, such as the variation of temperature, humidity and stress (i.e. of constraints exerted on a material), and on the other hand into so-called "slow" effects, exhibiting a variation on a time scale of the order of months, years, etc., such as the aging of sensors, the modification of material properties, of the structure or of the coupling between sensors and structure, and due to thermomechanical or other cycles.

[0013] In practice, the signatures of early defects are much smaller than the perturbations in measurements associated with the presence of such epistemic uncertainties and / or the presence of such unsupervised influencing effects. In other words, guided waves, particularly elastic and ultrasonic ones, used to detect structural defects via their signature, are not only sensitive to defects, but also to external and operational effects such as temperature, humidity, sensor uncertainties, aging, etc.

[0014] A classic solution aims to compensate for epistemic uncertainties by means of a reference state, i.e. a measurement in the absence of a fault and then comparing the current state to the reference state, in particular by subtraction, correlation, etc., while assuming that the only difference between the reference and current states can only be attributed to the presence of a fault, but that all other operating parameters, apart from at best a single fast-acting parameter such as temperature, are equal between these two reference and current states.

[0015] To compensate for rapid effects, such as a temperature variation, signal processing is generally used.

[0016] Such a classical solution is limiting, often prohibitive in terms of experimental design rimental, and likely to lead to the detection of a false positive or a false negative. Indeed, such a state-of-the-art solution first implies "perfect" instrumentation of the structure before the presence of a defect under identical operating conditions or at best up to a factor such as temperature.

[0017] Temperature compensation is also limited in terms of the maximum compensable range, which is currently on the order of a maximum temperature difference of 15°.

[0018] Moreover, effects other than temperature, such as external forces, are more difficult to reproduce in the laboratory, and the simultaneous presence of several effects has not really been explored until now, as well as the evolution of the reference state over time, a reference state which is likely to change despite the absence of a defect.

[0019] In other words, conventional solutions do not take into account the aforementioned slow effects, and are implemented under the assumption that the system behaves nominally over long periods, which is of course not satisfactory when considering the instrumentation of a structure over several decades during which the aging of the measurement sensor and its coupling to the structure is inevitable and generates a slow drift of the signals received by the measurement sensor, which may be wrongly interpreted as a defect, or may conceal one.

[0020] The aim of this invention is therefore to propose a method and a device suitable for providing an early and reliable diagnosis of structural defects under real conditions of use in the presence of epistemic uncertainties and unsupervised external effects both with rapid dynamics (i.e. on the order of hours to several days) and slow dynamics (i.e. on the order of months to several years).

[0021] To this end, the invention relates to a method for detecting anomaly(ies) in a structure, said structure carrying at least one sensor configured to generate and receive guided waves, the method comprising at least one learning phase and at least one operational phase,

[0022] the learning phase comprising the following steps:

[0023] - obtaining a set of healthy data representative of N healthy states of said structure respectively associated with N sets, distinct in pairs, of conditions of use of said structure, N being an integer, two distinct sets presenting at least one condition of use distinct from one set to another;

[0024] - projection of said healthy data set into a latent space of dimension reduced in relation to the size of said healthy dataset;

[0025] - determination of the outline of said healthy data set projected in said latent space;

[0026] the operational phase comprising the following steps:

[0027] - obtaining, by prior measurement via said at least one sensor, a set of test data representative of the current state of said structure;

[0028] - projection of said test data set into said latent space;

[0029] - detection of at least one common anomaly of said structure as soon as an element said test dataset is outside said outline.

[0030] This method aims to quantify the extent to which signal data, corrupted over various timescales, can be used for structural state analysis. Such a method thus makes it possible to compensate for the influence of the aforementioned supervised and / or unsupervised external effects without erasing the signature of the small defects that one seeks to detect.

[0031] More specifically, the invention consists of obtaining all the signal data that can be measured under healthy real-world conditions, said data being hereafter referred to as healthy data, that is, in the absence of defects, with the aim of detecting a defect by its absence from this set. In other words, obtaining this data amounts to constructing a digital twin describing all the healthy signals and then detecting an anomaly as being beyond the boundary of the "healthy" space described by this twin. Such a "healthy" space obtained according to the present invention is an important and differentiating element compared to conventional solutions. Indeed, according to the present invention, there is no need to model defects, as these will be detected by their absence from this set of healthy data. Not modeling defects is a significant advantage because it means not making assumptions related to the type, position, or size of the defects.

[0032] Since there are an infinite number of defect-free signals, the present invention further proposes to use a reduction of dimensionality, via a latent space, to describe the characteristics of signals in a healthy state.

[0033] According to other advantageous aspects of the invention, the method for detecting anomaly(ies) in a structure comprises one or more of the following features, taken individually or in all technically possible combinations:

[0034] - said healthy dataset is obtained following a prior phase of ca libration of said structure for the N distinct pairs of conditions of use of said structure;

[0035] - said healthy dataset is obtained following a prior phase of si mulation of said structure for the N distinct pairs of conditions of use of said structure;

[0036] - said healthy dataset is obtained following a preliminary hybrid phase of calibration and / or simulation of said structure for the N sets, distinct in pairs, of conditions of use of said structure;

[0037] - the process includes, during said hybrid preliminary phase, a step of com learning by transfer in case of discrepancy between calibration and simulation for the same set of conditions of use of said structure;

[0038] - said obtaining of a healthy dataset is implemented via a plurality of Q measurement sensors carried by said structure forming a sensor network, Q being an integer greater than one;

[0039] - said reduced-dimensional latent space is obtained by dimensional reduction su supervised or unsupervised;

[0040] - said unsupervised dimensional reduction is implemented by means of a elements belonging to the group comprising at least:

[0041] - a principal component analysis,

[0042] - a pre-trained autoencoder for compressing and then decompressing the signals from the healthy dataset, and of which only the part dedicated to compression is used to implement said dimensionality reduction.

[0043] - an autoregressive process;

[0044] - said supervised dimensional reduction is implemented by means of a network of neurons;

[0045] - said neural network is a neural network whose type belongs to the group including:

[0046] - a convolutional neural network;

[0047] - a recurrent neural network;

[0048] - a multilayer perceptron;

[0049] - said contour determination consists of:

[0050] - search, in said latent space, for the spherical or elliptical envelope of radius(s) minimal(s) encompassing the points of said healthy dataset projected into said latent space, or

[0051] - search in the latent space of the hyperplane furthest from the origin which separates, from the origin, the points of said healthy data set projected into said latent space, or

[0052] - use other types of anomaly detection such as robust estimation of the covariance matrix, known isolation forest, or even with outlier detection with a local outlier factor.

[0053] The invention also relates to a computer program comprising software instructions which, when executed by a computer, implement a method for detecting anomaly(ies) in a structure as defined above.

[0054] The invention also relates to an electronic device for detecting anomaly(ies) in a structure, said structure carrying at least one sensor for measuring at least one characteristic of said structure, the device comprising a unit learning and a test unit,

[0055] the learning unit comprising:

[0056] - a first retrieval module configured to obtain a dataset healthy representative of N healthy states of said structure respectively associated with N pairs of distinct sets of conditions of use of said structure, N being an integer, two distinct sets presenting at least one distinct condition of use from one set to another;

[0057] - a first projection module configured to project said dataset healthy in a latent space of reduced dimension compared to the dimension of said healthy data set;

[0058] - a determination module configured to determine a contour of said assembly of healthy data projected into said latent space;

[0059] the test unit comprising:

[0060] - a second obtaining module configured to obtain, by measurement via said at minus one sensor, a set of test data representative of the current state of said structure;

[0061] - a second projection module configured to project said set of test data in said latent space provided by the projection module of said learning unit;

[0062] - a detection module configured to detect at least one common anomaly of said structure as soon as an element of said test dataset is outside said outline provided by said learning unit determination module.

[0063] The invention also relates to a system for detecting anomaly(ies) in a structure, said system comprising the electronic device for detecting anomaly(ies) in a structure referred to herein and a plurality of Q measurement sensors carried by said structure forming a sensor network, Q being an integer greater than one.

[0064] According to a particular aspect of the aforementioned anomaly detection system in a structure, at least one of said Q sensors is configured to generate and receive guided ultrasonic elastic waves.

[0065] These features and advantages of the invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the accompanying drawings, in which:

[0066] [Fig-1] [Fig. 1] illustrates the effect of a structural defect on the propagation of guided waves on the surface of a structure.

[0067] [Fig.2] [Fig.2] is a schematic view of an electronic detection device of anomaly(ies) in a structure according to the present invention;

[0068] [Fig.3] [Fig.3] is a flowchart of a method for detecting anomaly(ies) in a structure according to the present invention;

[0069] [Fig.4][Fig.5] Figures 4 and 5 illustrate respectively the learning and the operational phase of the process according to the present invention.

[0070] In [Fig. 1], an example of guided wave propagation on the surface of a structure 10 is illustrated for the structure without a defect in view A and in the presence of a defect in view B.

[0071] More specifically, in view A of this example, the defect-free structure 10 corresponds to a defect-free cylindrical pipe of length L, carrying on its surface at least one measurement sensor in an uncontrolled environment, for example configured to generate, via element 12, and receive via element 14, guided waves 16, in particular elastic and ultrasonic waves. For example, such a sensor belongs to the group comprising at least: piezoelectric sensors, an EMAT (Electromagnetic Acoustic Transducer), a PVDF (Polyvinylidene Fluoride) sensor, etc. According to hybrid instrumentation, reception via element 14 can also be ensured by one of the Bragg gratings on optical fiber.

[0072] Other types of sensors are suitable for use provided they are capable of performing measurements in an uncontrolled environment, for example, sensors based on the use of ultrasound in general, including, as mentioned above, sensors capable of generating and receiving guided waves, but also eddy current or vibration measurement sensors, etc. In view B, the presence of a defect 18 on the surface of the pipe modifies the propagation of guided waves by generating, in particular, reflected guided waves 20.

[0073] Figure 2 is a schematic view of an electronic device 30 for detecting anomaly(ies) in a structure according to the present invention. Such a device 30 comprises a machine learning unit 32 including a first acquisition module 34 configured to obtain a set of healthy data representative of N healthy states of said structure respectively associated with N pairs of distinct sets of usage conditions of said structure, N being an integer, two distinct sets having at least one usage condition distinct from one set to another.

[0074] The learning unit 32 further includes a first projection module 36 configured to project said healthy dataset (also called healthy base comprising N samples, each sample being of dimension P) into a latent space E of reduced dimension compared to the dimension of said healthy dataset.

[0075] According to an optional aspect, such a latent space E is configured to describe the aforementioned disturbing effects as simply as possible, and is ideally linear, if not monotonous.

[0076] The learning unit 32 also includes a determination module 38 configured to determine a contour C, in particular multidimensional, of said healthy data set projected into said latent space.

[0077] Furthermore, the electronic device 30 includes a test unit 40, adapted to receive as input the output S of the learning unit 32, and comprising a second acquisition module 42 configured to obtain, by measurement via said at least one sensor, a set of test data representative of the current state of said structure 10.

[0078] The test unit 40 further includes a second projection module 44 configured to project said test data set into said latent space provided by the first projection module 36 of said learning unit 32.

[0079] The test unit 40 also includes a detection module 46 configured to detect at least one common anomaly of said structure as soon as an element of said test dataset is outside said contour provided by said determination module 38 of said learning unit 32.

[0080] In other words, the learning unit 32 is capable of constructing by training a digital twin describing the set of healthy data in order to detect an anomaly as beyond the healthy space described by this digital twin.

[0081] In the example of [Fig.2], the electronic device 30 for detecting anomaly(ies) in a structure includes an information processing unit 50 formed for example of a memory 52 and a processor 54 associated with the memory 52.

[0082] In the example of [Fig.2], the first acquisition module 34, the first projection module 36, the determination module 38, the second acquisition module 42, the second projection module 44 and the detection module 46 are each implemented as a software program, or a software block, executable by the processor 54. The memory 52 of the electronic device 30 for detecting anomaly(ies) in a structure is then capable of storing, for the implementation of a learning phase, a first acquisition software program, a first projection software program, and a determination software program, and for the implementation of a test phase, a second acquisition software program, a second projection software program, and a detection software program.The processor 54 is then capable of executing each of the following software programs: the first acquisition software, the first projection software, the determination software for a learning phase, and for a testing phase, a second acquisition software, a second projection software, and a detection software.

[0083] In an alternative not shown, the first acquisition module 34, the first projection module 36, the determination module 38, the second acquisition module 42, the second projection module 44 and the detection module 46 are each carried out in the form of a programmable logic component, such as an FPGA (Field Programmable Gate Array), or a GPU (Graphics Processing Unit), or in the form of an integrated circuit, such as an ASIC (Application Specified Integrated Circuit).

[0084] When the electronic device 30 for detecting anomaly(ies) in a structure is implemented in the form of one or more software programs, i.e., in the form of a computer program, also called a computer program product, it is further capable of being stored on a computer-readable medium (not shown). A computer-readable medium is, for example, a medium capable of storing electronic instructions and being connected to a bus of a computer system. For example, a readable medium is an optical disc, a magneto-optical disc, a ROM, a RAM, any type of non-volatile memory (e.g., EPROM, EEPROM, FLASH, NVRAM), a magnetic card, or an optical card. A computer program comprising software instructions is then stored on the readable medium.

[0085] Not shown, the invention also relates to a system for detecting anomaly(ies) in a structure, said system comprising such an electronic device 30 for detecting anomaly(ies) in a structure, illustrated by [Fig. 2], and further comprising a plurality of Q sensors carried by said structure, configured to generate and receive guided waves and forming a sensor array, Q being an integer greater than one. According to a particular aspect, at least one of said Q sensors is configured to generate and receive guided ultrasonic elastic guided waves.

[0086] In this case, the information is therefore processed at the sensor network level by exploiting Q2-Q signals corresponding to the paths between the Q sensors, or Q2 signals if the signals emitted and received by the same sensor are used. The sensor network Q is optionally supplemented by an acquisition chain (not shown) configured to process the measurements captured via said network of Q sensors.

[0087] Indeed, it is possible that a signal acquired by a sensor element 14, as illustrated in [Fig. 1], in the presence of a fault under operating conditions U1, might be identical to a signal acquired in the absence of a fault under conditions U2. To resolve this ambiguity, the present invention proposes to base the analysis on data from a sensor network. In fact, due to the asymmetry of faults, the contribution of a given fault varies according to the paths traveled by the waves, which is not the case for operating conditions. The data is therefore processed at the sensor network level, with Q2-Q signals corresponding to the paths between the Q sensors.

[0088] The operation of the electronic device 30 for detecting anomaly(ies) in a structure will now be described with reference to [Fig. 3], which illustrates the diagram. ticement an example of implementation, according to the present invention, of a method 60 for detecting anomaly(ies) in a structure, such as for example structure 10 of [Fig.1].

[0089] The method 60 according to the present invention first includes a machine learning phase 62, also illustrated in more detail in [Fig.4] described below.

[0090] According to a first step 64 of the learning phase 62, the electronic device 30 for detecting anomaly(ies), via its first acquisition module 34, obtains OBT_A a set of healthy data representative of N healthy states of said structure respectively associated with N sets, distinct two by two, of conditions of use of said structure, N being an integer, two distinct sets presenting at least one condition of use distinct from one set to another.

[0091] The N sets, distinct in pairs, of conditions of use of said structure are intended in particular to cover the main operating parameters likely to influence the measurements, namely those associated with the aforementioned epistemic effects, such as the placement of sensors, the properties and geometry of the structure, the properties of the sensors, those associated with the aforementioned rapid effects, such as temperature, humidity, pressure, the forces applied to the structure in question, those associated with slow effects such as the aging of the sensor(s), of the structure and of the coupling between the structure and the sensor(s).

[0092] As illustrated by [Fig.2], such an attainment 64 follows one of the three optional preliminary phases 642, 644, 646, implemented by said electronic anomaly detection device 30 or by a separate device, the first optional preliminary phase 642 corresponding to a preliminary calibration phase CAL of said structure for the N distinct pairs of conditions of use of said structure.

[0093] In other words, according to this first option, the healthy dataset is determined beforehand through experience during a calibration phase 642, which is often prohibitively expensive in terms of cost / time, and consists of instrumenting healthy structures and acquiring data over long periods. Such a calibration phase 642 involves both monitoring the measured data to ensure that there are no defects, and ensuring its completeness so that a statistically sufficient number of parameter combinations are explored. Depending on the application, such calibration is available and performed beforehand, for example, during a research and development phase of the structure in question, with the acquisition step 64 then consisting of retrieving the healthy dataset from the calibration phase 642.

[0094] According to a second preferred option, said healthy data set is obtained at the end of a preliminary phase 644 of SIM simulation of said structure for the N distinct pairs of conditions of use of said structure, the obtaining step 64 then consisting of recovering the set of healthy data from the phase 644 of SIM simulation.

[0095] This second option is a preferred approach, according to the present invention, the prior simulation 644 allowing to establish, quickly and at a lower cost than the first experimental option, models (i.e. simulations) of the signals subjected to the disturbing parameters corresponding to the epistemic uncertainties and the aforementioned unsupervised external effects, and thus to generate the healthy data on the fly.

[0096] Indeed, depending on the application, it may prove prohibitive in the first instance to acquire all possible healthy signals experimentally, in particular on the one hand because the effects of aging are only visible after several years, and on the other hand because it would be necessary to instrument many structures to observe all the variabilities of structures / instrumentation.

[0097] Such a second simulation option also implies having available models of all the phenomena influencing guided waves, such models being able to be learned / calibrated / tested / validated on representative samples of the structure and application under consideration, greatly limiting the cost compared to the experimental solution associated with the first option mentioned above. Advantageously, defect models (i.e., simulations) are not necessary, since the present invention proposes to focus on the healthy data.

[0098] According to a third option, said healthy data set is obtained following a prior hybrid calibration and / or simulation phase 646 of said structure for the N pairs of distinct sets of conditions of use of said structure, the obtaining step 64 then consisting of recovering the healthy data set from the hybrid HYB phase 646.

[0099] According to a particular optional aspect of this third option, the aforementioned hybrid preliminary phase includes an unshown step of compensation by transfer learning in case of discrepancies between calibration and simulation for the same set of operating conditions of said structure. In other words, such an optional aspect is implemented in particular in the case where the models (i.e., simulations) are not perfect, that is to say, without rigorous superposition between the simulated and experimental signals with identical parameters, and proposes to compensate for the discrepancy by transfer learning, an example of which is described in French patent application FR 3 113 530, such "transfer learning" allowing, in a hybrid manner, the recalibration of the models using a manageable number of experiments.

[0100] Regardless of the option implemented prior to obtaining step 64, As previously mentioned, it should be noted that for each option, an anomaly detection system is considered for a structure. This system comprises the electronic anomaly detection device 30, illustrated in [Fig. 2], and further comprises a plurality of Q sensors carried by said structure, configured to generate and receive guided waves and forming a sensor network, where Q is an integer greater than one, and N is such that N = Q² - Q. The network of Q sensors ensures that the dataset retrieved during the acquisition step 64 is sound.

[0101] Indeed, external effects are relatively homogeneous and perceived by all sensors. This is the case, for example, with temperature, which tends to affect all measuring sensors in a relatively homogeneous manner. On the other hand, a defect will only influence certain sensors and in quite different ways, due to its limited size and natural asymmetry.

[0102] Furthermore, regardless of the option implemented prior to the acquisition step 64, the learning phase 62 of the method 60 according to the present invention further comprises a projection step 66 of said healthy data set into a latent space E of reduced dimension compared to the dimension of said healthy data set. Indeed, the healthy data set is generally of high dimension, in particular on the order of 10³ to 10⁵, which tends to slow down anomaly detection during the operational testing phase.

[0103] The projection step 66 therefore aims to reduce the dimensions of the healthy training data, by describing them (i.e. projecting) into a reduced dimension latent space, for example of the order of 10 to 102 for original dimensions respectively of 103 to 105.

[0104] Said reduced-dimensional latent space is obtained by supervised or unsupervised dimensionality reduction

[0105] It should be noted that since the healthy data do not contain any defects, an unsupervised dimensional reduction avoids introducing any bias against defect detection. In other words, an unsupervised dimensional reduction is appropriate for implementing in order to avoid creating a bias that would contradict the signature of defects, which are absent from the initial set of healthy data.

[0106] According to a first option, when the dimensional reduction 66 is unsupervised, it is implemented by means of one of the elements belonging to the group comprising at least:

[0107] - a principal component analysis PCA (from the English Principal Component Analysis),

[0108] - a pre-trained autoencoder for compressing and then decompressing the signals from the healthy dataset, of which only the part dedicated to the com- Pressure is used to implement said dimensional reduction.

[0109] - an autoregressive process

[0110] In other words, the first projection module 36 is designed to implement principal component analysis, ensuring in particular that the loss of variance is less than the influence of a defect subsequently sought during the operational phase. In the case of implementing an autoencoder, the first projection module 36 is configured to drive the first half of said autoencoder to compress / decompress the signals, and the dimensionality reduction is achieved by retaining only the central component of the autoencoder commonly called "embedding".

[0111] According to a second option, when the dimensional reduction 66 is supervised, in particular to best adapt a set of sound data from a prior simulation, whose physics is well controlled, and to promote the detection of defects, such a supervised dimensional reduction 66 is implemented by means of a neural network, the type of which belongs for example to the group comprising:

[0112] - a convolutional neural network;

[0113] - a recurrent neural network;

[0114] - a multilayer perceptron.

[0115] Such a neural network is particularly suited to modeling data and disturbing effects such as temperature, humidity, etc., the reduced dimension latent space then corresponding to the set of outputs of the neural network.

[0116] Furthermore, as illustrated by [Fig.2], the learning phase 62 of the method 60 according to the present invention further comprises a step 68 of contour determination C, implemented automatically via said determination module 38 or manually, of said healthy data set projected into said latent space E. In other words, during said step 68, the aim is, by training, to describe the contours of the healthy data set, in order to subsequently, during an operational phase, detect an anomaly outside said contour.

[0117] To this end, as described in particular by V. Chandola et al. in the publication entitled “Anomaly detection: A survey”, ACM Computing Surveys Volume 41 Issue 3 July 2009 Article No.: 15pp 1-58, said contour determination 68 is, for example, suitable for understanding:

[0118] - the search, in said latent space E, for the envelope, in particular spherical or el liptic, of minimal radius(s) encompassing the points of said healthy data set projected into said latent space, according to the technique known in English as "support vector data description", or

[0119] - the search in latent space E for the hyperplane furthest from the origin which separates, from the origin, the points of said healthy data set projected into said latent space, according to the technique known by the English name "one-class support vector machine", or

[0120] - use other types of anomaly detection such as robust estimation of the covariance matrix known by the English name "robust covariance", isolation forest known by the English name "isolation forest", or with outlier detection with the local outlier factor LOF known by the English name "local outlier factor".

[0121] Once the learning phase 62 is completed, the outputs S, comprising the latent space E and the contour C, of ​​the learning unit 32, are transmitted to the test unit 40, which is designed to implement an operational phase 70. This phase is repeated after each measurement step 72 M of a current state of the structure under consideration. As previously stated, each measurement 72 is obtained, in particular, using a plurality of Q sensors carried by said structure, configured to generate and receive guided waves and forming a sensor network, Q being an integer greater than one, and N being such that N = Q² - Q.

[0122] It should be noted that the learning phase 62 is likely to be repeated in particular to take into account additional parameters likely to influence the propagation of guided waves depending on the application of the desired structure, for example in the event of a change in the climatic zone of operation of said structure.

[0123] Furthermore, according to an aspect not shown, the present invention makes it possible to take into account the knowledge of a physical parameter of the condition of use at the time of the measurement 72, such a physical parameter corresponding in particular to the temperature, in order to reduce the set of healthy data from the learning 62 to a restricted set of healthy data of one dimension due to this known physical parameter.

[0124] In other words, in this case, the learning process 62 is repeated to obtain said restricted healthy dataset, and its contour C' in a latent space E' associated with this dimensionality reduction linked to perfect knowledge of a usage parameter such as the temperature during measurement 72. Such a contour C' is necessarily more restricted than the contour C associated with the healthy dataset obtained for a plurality of distinct operating temperatures of the structure. Such a more restricted contour C' makes the approach more reliable, and in the extreme case where all the influential parameters were known, which does not seem possible in practice due to epistemic uncertainties, the healthy dataset would then contain a single set of data.

[0125] Such an operational phase 70, also illustrated and described subsequently in relation to [Fig. 5], comprises a first step 74 of obtaining OBT_T, via the second obtaining module 42 of [Fig. 2], a set of test data represented sensations of the current state of said structure in question, in which the presence of a defect is unknown.

[0126] Then, during a step 76, implemented by the second projection module 44 of [Fig.2], the test data set representative of the current state of said structure under consideration is projected, via a P_T projection, into the latent space E, used during the learning phase 62.

[0127] Then, during a step 78, implemented automatically by the detection module 46 of [Fig.2] or manually, at least one current anomaly of said structure is detected as soon as an element of said test dataset is outside said contour C.

[0128] Not shown, the method 60 is also suitable for including an assessment of biases and errors due to modelling and machine learning in order to provide a measure of confidence in the fault diagnosis (i.e. anomaly detection) proposed according to the present invention.

[0129] Fig. 4 illustrates schematically, in more detail, the learning phase 62 previously described, with step 64 of obtaining a set of healthy data representative of N healthy states of said structure respectively associated with N pairs of distinct sets of conditions of use of said structure, N being an integer, two distinct sets having at least one distinct condition of use from one set to another.

[0130] More specifically, as illustrated in [Fig. 4], during this acquisition step 64, healthy data associated with as many signals N as there are distinct sets of operating conditions for a healthy, defect-free structure are obtained. In other words, each of the N signals shown is obtained with a variation of at least one operating parameter such as the position of at least one sensor, the temperature, and / or the aging of said at least one sensor.

[0131] Step 66 of projection of said healthy data set into a latent space of reduced dimension M with respect to the dimension N of said healthy data set is illustrated in [Fig.4] by M2 two-dimensional representations of the healthy data represented by crosses.

[0132] It should be noted that such two-dimensional representations are used here solely for illustrative and explanatory purposes of the multidimensional aspect of latent space, the multidimensional representation of which is complex.

[0133] Step 68 of determining the contour of said healthy data set (also called the healthy base comprising N samples, each sample being of dimension P) projected into said latent space is illustrated in [Fig. 4] by the contours Ci and C2, the contour Ci surrounding the healthy data crosses projected in step 66 into a two-dimensional part with dimension 1 on the abscissa and dimension 2 on the ordinate of the latent space of M dimensions (i.e., of size M) <P nécessairement du fait de la réduction de dimension), le contour C2 entourant les croix de données saines projetée lors de l’étape 66 dans une partie bidimensionnelle avec la dimension 1 en abscisse et la dimension 3 en ordonnée de l’espace latent à M dimensions.

[0134] Figure 5 illustrates schematically, in more detail, the operational phase previously described, with step 74 of obtaining, by prior measurement via said at least one sensor, a set of test data representative of the current state of said structure.

[0135] The step 76 of projection of said test data set into said latent space, from the training 62 illustrated by [Fig.4], is represented on [Fig.5] by M2 two-dimensional representations of the test data Ti and T2 represented by a cross in each two-dimensional representation.

[0136] Anomaly detection step 78 compares, in each two-dimensional representation, whether the test data cross is within the associated healthy data contour. Otherwise, an anomaly 82 is detected as soon as a cross is outside a two-dimensional contour. This is the case for cross Ti, which is outside contour Ci in the two-dimensional space of dimensions 1 and 2, whereas when dimensions 1 and 3 are considered, cross T2 is within contour C2. In other words, as soon as a test data position outside the multi-dimensional contour is detected, particularly here for explanatory purposes and simplicity of representation within the M2 two-dimensional representations, an anomaly is detected, and if applicable, in an unrepresented manner, an unrepresented alert is raised and / or a maintenance request is issued for said structure.

[0137] In other words, such a detection 78 is admittedly crude but effective, and consists of considering that if the representation in the latent space of the test data set is too far from the contour C of the healthy data set, a defect is present.

[0138] According to a particular aspect, very dependent on the use case considered, the distance of the test data representative of an anomaly (here the cross TJ with the contour here Ciperrait could be specific to reflect the criticality of the detected defect.

[0139] A person skilled in the art will understand that the invention is not limited to the embodiments described, nor to the particular examples of the description, the embodiments and variants mentioned above being capable of being combined with each other to generate new embodiments of the invention.

[0140] Thus, the present invention proposes a method and a device for detecting anomaly(ies) in a structure which makes it possible to build, by training, a digital twin describing the set of healthy data associated with a healthy state of the structure considered in order to then detect an anomaly as beyond the healthy space described by this digital twin.

[0141] In other words, the present invention makes it possible to avoid the creation of defects whether experimentally or by simulation, by learning through the modeling of only healthy states.

[0142] According to one aspect of the invention implementing the obtaining of sound data at least partly by prior simulation, it is possible to do away with a prior calibration phase on real structures and under real conditions of use, or to greatly limit it in the case of a hybrid prior processing involving at least partly simulation.

[0143] Moreover, the present invention is generic with respect to influential parameters, because once the healthy dataset is obtained by learning, it is possible to evolve it rapidly to take into account additional conditions of use of the structure and recalculate its multidimensional contour in latent space.

[0144] Thus, the anomaly detection obtained via the present invention exhibits an intrinsic robustness to external parameters by integrating them from the outset into the learning of healthy data, so that the diagnosis (i.e. the detection) does not make any a priori assumptions about external conditions at the time of its application to the measurement of a current state of the structure.

[0145] Such anomaly detection is likely to reduce the environmental footprint of instrumented structures through preventive maintenance and extending the lifespan by detecting anomalies (i.e. defects) at an early stage.

Claims

Demands

1. A method (60) for detecting anomaly(ies) in a structure and for SHM health monitoring of said structure, an anomaly corresponding to modifications of the physical and / or geometric properties of the structure, said structure having at least one sensor for measuring at least one characteristic of said structure, the method comprising at least one learning phase (62) and at least one operational phase (70), the learning phase (62) comprising the following steps: - obtaining (64) a set of healthy data representative of N healthy states of said structure respectively associated with N pairs distinct sets of conditions of use of said structure, N being an integer, two distinct sets having at least one condition of use distinct from one set to another,said acquisition of a healthy dataset being implemented via a plurality of Q measurement sensors carried by said structure forming a sensor network, Q being an integer greater than one, at least one of said Q sensors being configured to generate and receive guided ultrasonic elastic waves; - projection (66) of said healthy data set into a latent space of reduced dimension compared to the dimension of said healthy data set; - determination (68) of the outline of said healthy data set projected into said latent space; the operational phase (70) comprising the following steps: - obtaining (74), by prior measurement (72) via said at least one sensor, a set of test data representative of the current state of said structure; - projection (76) of said test data set into said latent space; - detection (78) of at least one current anomaly of said structure as soon as an element of said test dataset is outside said contour.

2. Method (60) according to claim 1, wherein said healthy data set is obtained following a prior calibration phase (642) of said structure for the N distinct pairs of usage conditions of said structure.

3. Method (60) according to claim 1, wherein said healthy data set is obtained following a prior simulation phase (644) of said structure for the N pairwise distinct sets of conditions of use of said structure.

4. Method (60) according to claim 1, wherein said healthy data set is obtained following a prior hybrid calibration and / or simulation phase (646) of said structure for the N pairs distinct sets of conditions of use of said structure.

5. Method (60) according to claim 4, comprising, during said hybrid preliminary phase, a transfer learning compensation step in case of discrepancy between calibration and simulation for the same set of operating conditions of said structure.

6. Method (60) according to any one of the preceding claims, wherein said reduced dimension latent space is obtained by supervised or unsupervised dimensional reduction.

7. A method (60) according to claim 6, wherein said unsupervised dimensionality reduction is implemented by means of one of the elements belonging to the group comprising at least: - a principal component analysis, - an autoencoder pre-trained to compress and then decompress the signals from the healthy dataset, and of which only the part dedicated to compression is used to implement said dimensionality reduction, - an autoregressive process.

8. Method (60) according to claim 6, wherein said supervised dimensional reduction is implemented by means of a neural network.

9. Method (60) according to claim 8, wherein said neural network is a neural network whose type belongs to the group comprising: - a convolutional neural network; - a recurrent neural network; - a multilayer perceptron.

10. A method (60) according to any one of the preceding claims, wherein said contour determination (68) consists of: - searching, in said latent space, for the spherical or elliptical envelope of minimum radius(s) encompassing the points of said healthy data set projected into said latent space, or

11.

12. - search in the latent space of the hyperplane furthest from the origin that separates, from the origin, the points of said healthy data set projected into said latent space, or - use other types of anomaly detection such as robust estimation of the covariance matrix, known isolation forest, or outlier detection with a local aberration factor. A computer program comprising software instructions which, when executed by a computer, implement, at least in part, a method for detecting anomaly(ies) in a structure according to any one of the preceding claims. A system for detecting anomaly(ies) in a structure and for monitoring the health of said structure, an anomaly corresponding to changes in the physical and / or geometric properties of the structure, said system comprising a plurality of Q sensors for measuring at least one characteristic of said structure, carried by said structure and forming a sensor array, Q being an integer greater than one, at least one of said Q sensors being configured to generate and receive guided ultrasonic elastic waves, said system comprising a device (30) in said structure and comprising a learning unit (32) and a test unit (40), the learning unit (32) comprising: - a first acquisition module (34) configured to obtain, via said plurality of Q measurement sensors, a set of healthy data representative of N healthy states of said structure respectively associated with N pairs of distinct sets of conditions of use of said structure, N being an integer, two distinct sets having at least one distinct condition of use from one set to another; - a first projection module (36) configured to project said set of healthy data into a latent space of reduced dimension compared to the dimension of said set of healthy data; - a determination module (38) configured to determine an outline of said healthy dataset projected into said latent space; the test unit (40) comprising: - a second acquisition module (42) configured to obtain, by measurement via said at least one sensor, a set of test data representative of the current state of said structure; - a second projection module (44) configured to project said test dataset in said latent space provided by the projection module of said learning unit; - a detection module (46) configured to detect at least one common anomaly of said structure as soon as an element of said test dataset is outside said contour provided by said determination module of said learning unit.

Images