WO2020221618A1

WO2020221618A1 - Method for detecting soft faults in a cable by analysing main components

Info

Publication number: WO2020221618A1
Application number: PCT/EP2020/061048
Authority: WO
Inventors: Nour TAKI; Wafa BEN HASSEN; Nicolas Ravot; Claude DELPHA; Demba DIALLO
Original assignee: Commissariat A L'energie Atomique Et Aux Energies Alternatives; Centralesupelec; Centre National De La Recherche Scientifique; Sorbonne Universite; Universite Paris-Sud
Priority date: 2019-04-30
Filing date: 2020-04-21
Publication date: 2020-11-05
Also published as: FR3095700B1; FR3095700A1

Abstract

A method for detecting faults in a transmission line (L) comprising the following steps: - defining (300) a plurality of test signals each having a different maximum spectral frequency band; - applying (302) an analysis of main components to the first multidimensional variable in order to determine a model of the transmission line at an initial instant; - acquiring (303) a plurality of temporal reflectograms corresponding to test signals previously introduced into the line (L) at a current instant; - determining (304) at least one prediction error; - comparing (305) the at least one prediction error to a fault detection threshold and deducing therefrom the existence of a fault when at least one prediction error exceeds a fault detection threshold for at least one sample of the multidimensional variable.

Description

Title of the invention: Method for detecting non-clear faults in a cable by principal component analysis

The invention relates to the field of wired diagnostic systems based on the principle of reflectometry. It relates to a method for detecting non-blunt faults in a cable using several test signals at different frequencies and a principal component analysis.

An objective of the invention is to exploit a diversity in frequency of the test signals to evaluate the best choice of frequency according to the characteristics of the cable and the fault to be detected.

Cables are ubiquitous in all electrical systems, whether for power or for transmitting information. These cables are subject to the same stresses as the systems they connect and can be subject to failure. It is therefore necessary to be able to analyze their condition and provide information on fault detection, but also their location and type, in order to help with maintenance. The usual reflectometry methods allow this type of test.

The reflectometry methods use a principle similar to that of radar: an electrical signal, the probe signal or reference signal, which is most often of high frequency or wide band, is injected into one or more places of the cable to be tested. The signal travels through the cable or network and returns part of its energy when it encounters an electrical discontinuity. An electrical discontinuity can result, for example, from a connection, from the end of the cable or from a fault or more generally from a rupture of the signal propagation conditions in the cable. It most often results from a fault which locally modifies the characteristic impedance of the cable by causing a discontinuity in its linear parameters. Analysis of the signals returned to the injection point makes it possible to deduce therefrom information on the presence and location of these discontinuities, and therefore of any faults. Usually, time or frequency domain analysis is performed. These methods are designated by the acronyms TDR coming from the Anglo-Saxon expression “Time Domain Reflectometry” and FDR coming from the Anglo-Saxon expression “Frequency Domain Reflectometry”.

The invention falls within the scope of reflectometry methods for wire diagnostics and applies to any type of electric cable, in particular power transmission cables or communication cables, in fixed or mobile installations. . The cables concerned can be coaxial, bifilar, in parallel lines, in twisted pairs or other provided that it is possible to inject a reflectometry signal into a point of the cable and to measure its reflection at the same point or at another point. .

The TDR and FDR reflectometry methods as well as the methods derived from the latter such as the MCTDR (Multi-Carrier Time Domain Reflectometry) or OMTDR (Orthogonal Multi-tone Time Domain Reflectometry) method have proven their effectiveness in the detection and localization of clear faults, that is to say faults which significantly impact the local characteristic impedance of a cable, for example faults associated with deterioration of the dielectric material constituting the cable.

On the contrary, when it comes to detecting and locating the presence of non-blunt defects, that is to say superficial defects, these methods are relatively limited. This difficulty is due to the fact that a non-blunt fault results in a very small variation in characteristic impedance and a reflection coefficient which is also very low. Consequently, a signal reflecting off such a non-blunt fault will generate a reflection of very low amplitude compared to the amplitude of an identical reflection on a blunt fault. The peak amplitude, also called signature, of the non frank can therefore be drowned in the measurement noise or masked by an adjacent frank fault.

A non-clear fault generally introduces a local modification of the electrical characteristics of a transmission line, for example of a cable. The physical degradation, even superficial, of the line leads in particular to a local modification of the characteristic impedance of the cable, which results in the modification of the reflection coefficient at the location of the fault.

The term non-blunt defect refers here to any defect having a surface impact on a cable so as to generate a variation in the characteristic impedance locally. In particular, such defects include scratching or wear of the sheath, of the dielectric, but also the onset of degradation of the metallic conductor, compression of a cable, friction or even corrosion. These degradations may, at first glance, appear benign and without significant repercussions for the system. However, if nothing is done, the mechanical and environmental constraints or even the aging of the cable, will cause a non-obvious fault to evolve into a clear fault, the consequences of which, both economic and material, can be considerable. Detecting emerging faults allows better management of maintenance and therefore a reduction in repair costs.

Moreover, an important parameter for the time reflectometry methods concerns the choice of the frequency bandwidth of the test signal used. Indeed, this parameter has a direct impact on the ability to detect a non-blunt fault as well as on the detection precision. On the one hand, when the duration of the test signal pulse is greater than the time it takes for the signal to pass through the fault (this duration being linked to the length of the fault), the fault signature can be made invisible on reflectogram. Thus, a test signal having a Too narrow frequency band does not make it possible to detect defects of too large a width. Conversely, a test signal having a frequency band that is too high has the drawback of a greater attenuation of the signal, in particular for cables of great length. Thus, the detection of an echo in the reflected signal is made more difficult because of this attenuation.

It can therefore be seen that a compromise must be found to select the most optimal frequency band for the test signal according to the characteristics of the cable to be analyzed and the type of fault to be detected.

Typically, implementing a time reflectometry method involves injecting into the cable a single test signal having a given frequency bandwidth. If the bandwidth of the test signal is not suitable for the faults present on the cable, they will not be detected.

There is therefore a need for a method making it possible to adapt the passband of the test signal to take into account the variability of the characteristics of the cables and of the faults to be analyzed so as to improve the detection of faults of various kinds, in particular faults. not frank.

The literature contains various publications describing methods for detecting defects by reflectometry for cables or cable networks. Reference may be made in particular to references [1] and [2]. More particularly, methods suitable for the detection of non-frank defects are described in references [3] - [6].

However, neither of these methods addresses the issue of which test signal bandwidth to use.

The document [7] describes a method which combines a reflectometry approach and an artificial neural network to solve a problem inverse which consists in detecting faults from the temporal reflectogram.

Overall, the state of the art does not contain a solution to adaptively select the bandwidth of the test signal according to the characteristics of the cable and the faults to be analyzed.

The invention proposes a method based on a principal component analysis of a multidimensional variable composed of several reflectograms resulting from reflectometry tests with signals of different bandwidths.

The invention makes it possible, on the one hand, to improve the ability to detect a fault by jointly exploiting several reflectometry measurements at different frequencies. On the other hand, the invention makes it possible to select the most optimal frequency band for the continuous monitoring of the condition of a cable or for the analysis of a cable of the same type as that used to implement the cable. 'invention.

The subject of the invention is a method for detecting non-clear faults in a transmission line comprising the steps of:

- Define several test signals each having a different maximum frequency spectral band, all of the test signals forming a first multidimensional variable,

- Apply to the first multidimensional variable, a principal component analysis to determine a model of the transmission line from the measurements of signals characteristic of the reflection of said test signals previously injected into the line at an initial instant, each measurement being called time reflectogram,

- Acquire several time reflectograms corresponding to said test signals previously injected into the line at a current instant so as to form a second multidimensional variable,

- Determine at least one prediction error between the second multidimensional variable and a projection of the second multidimensional variable on a space defined by the model of the transmission line at the initial time,

- Compare the at least one prediction error with a fault detection threshold and deduce the existence of a fault when at least one prediction error exceeds a fault detection threshold for at least one sample of the variable multidimensional.

According to a particular aspect of the invention, the space defined by the model of the transmission line at the initial time is a main subspace or a residual subspace defined by principal component analysis.

According to a particular aspect of the invention, the prediction error is taken from an error of the quadratic prediction error type or an error of the Hotelling statistical type T ² .

According to a particular aspect of the invention, the initial state of the line is a state for which the line exhibits no fault.

According to a particular variant, the method comprises the step of, for each sample of the multidimensional variable for which the prediction error exceeds a detection threshold, determining a relative contribution of each frequency to this sample.

According to a particular variant, the method comprises a step of selecting at least one frequency from among the frequencies which contribute the most to the samples for which a defect is detected.

According to a particular aspect of the invention, the defect detection threshold is a confidence threshold determined from the model obtained by principal component analysis.

According to a particular aspect of the invention, the test signals are time pulses of variable maximum frequency. According to a particular aspect of the invention, the test signals are multi-carrier signals of variable maximum frequency.

According to a particular aspect of the invention, the acquisition of a time reflectogram comprises the correlation of the measurement of the signal characteristic of the reflection of a test signal previously injected into the line and of a test signal.

The invention also relates to a device for analyzing a cable comprising an apparatus for measuring, at a point on the cable, a signal reflected in the cable and a computer configured to execute the method for detecting non-faulty faults. francs according to one of the preceding claims.

The subject of the invention is also a computer program comprising instructions for the execution of the method for detecting non-clear faults in a cable according to the invention, when the program is executed by a processor, as well as a support for 'recording readable by a processor on which is recorded a program comprising instructions for the execution of the method for detecting non-clear faults in a cable according to the invention, when the program is executed by a processor.

Other characteristics and advantages of the present invention will become more apparent on reading the following description in relation to the accompanying drawings which show:

- [Fig. 1] is a diagram illustrating the known principle of time reflectometry and its application to the detection of a non-clear fault,

- [Fig. 2] represents an example of reflectogram illustrating the appearance of the signature of a non-clear defect,

- [Fig. 3] is a flowchart detailing the steps in implementing a fault detection method according to the invention,

- [Fig. 4] represents an example of reflectometry measurements for a first type of fault, - [Fig. 4a] represents a diagram of an example of a fault on the sheath of a cable,

- [Fig. 5] represents an example of reflectometry measurements for a second type of fault,

- [Fig. 6] represents an example of reflectometry measurements for a third type of fault,

- [Fig. 7] represents a diagram of a first type of prediction errors obtained for the measurements in Figures 4,5 and 6,

- [Fig. 8] represents a diagram of a second type of prediction errors obtained for the measurements in Figures 4,5 and 6,

- [Fig. 9] represents a diagram of the contributions of the various measurements for a fault detected from the first type of prediction errors,

- [Fig. 10] represents a diagram of the contributions of the various measurements for a fault detected from the second type of prediction errors,

- [Fig. 1 1] shows an example of a fault detection device according to the invention,

Figure 1 shows schematically, as a reminder, the operating principle of a reflectometry diagnostic method applied to a transmission line L exhibiting a non-clear DNF fault. The example described below corresponds to a time reflectometry method.

A reference signal S is injected into the transmission line at a point P. The reflected signal R is measured at the same point P (or at another point on the line). This signal propagates in the line and encounters, during its propagation, a first impedance discontinuity at the input of the non-clear fault DNF. The signal is reflected on this discontinuity with a reflection coefficient G. If the characteristic impedance Z _c2 in the zone of the non-clear fault DNF is lower than the characteristic impedance Z _cl before the appearance of the fault, then the reflection coefficient G is negative and is reflected by a peak of negative amplitude in the reflected signal R. In the reverse case, the reflection coefficient G _c is positive and results in a peak of positive amplitude in the reflected signal R.

The transmitted part T of the incident signal S continues to propagate in the line and then encounters a second impedance discontinuity creating a second reflection of the incident signal with a reflection coefficient G ₂ of sign opposite to the first reflection coefficient 1 ^. If G <0 then G ₂ > 0. If G _c > 0 then G ₂ <0.

Thus, by observing the reflected signal R, the signature of the non-blunt defect DNF is characterized by two successive peaks with inverted signs as shown in Figure 2.

Figure 2 shows a time reflectogram which corresponds either directly to the measurement of the reflected signal R, or to the cross-correlation between the reflected signal R and the signal injected into the cable S.

In the case where the injected reference signal is a time pulse, which corresponds to the case of a time reflectometry method, the reflectogram can correspond directly to the measurement of the reflected signal R. In the case where the injected reference signal is a more complex signal, for example for MCTDR (Multi Carrier Time Domain Reflectometry) or OMTDR (Orthogonal Multi tone Time Domain Reflectometry) type methods, then the reflectogram is obtained by inter-correlating the reflected signal R and the injected signal S.

In FIG. 2, two reflectograms 201, 202 have been shown corresponding to two different pulse durations for the signal injected into the cable. Curve 201 corresponds to a pulse duration 2.DT much greater than the time for the signal to cross the non-clear fault DNF. The length of the fault being noted Ld, this duration is equal to Ld / V, with V the speed of propagation of the signal in the cable. Curve 202 corresponds to a pulse duration 2.DT much less than the time for the signal to pass through the non-clear fault DNF. In both cases, the signature 203 of the non-clear defect, in the reflectogram, is always composed of the succession of a first peak and a second peak, the signs of which are reversed.

The distance between the two peaks represents the length of the soft defect and their amplitude represents the severity of the soft defect. Indeed, the greater the variation of the characteristic impedance, the greater the amplitude of the signature of the non-blunt defect in the reflectogram is also important.

As is known in the field of reflectometry diagnostic methods, the position d _DNF of the faulty fault on the cable, in other words its distance from the signal injection point P, can be obtained directly from the measurement, on the temporal reflectogram of FIG. 2, of the duration ÎDNF between the first amplitude peak recorded on the reflectogram (at the abscissa 0.5 in the example of FIG. 3) and the amplitude peak 203 corresponding to the signature of the defect not clear.

Various known methods can be envisaged for determining the position d _DNF. A first method consists in applying the relationship between distance and time: dDNF = V.ÎDNF where V is the speed of propagation of the signal in the cable. Another possible method consists in applying a proportionality relation of the type doNF / ÎDNF = L / t ₀ where L is the length of the cable and to is the duration, measured on the reflectogram, between the amplitude peak corresponding to the discontinuity d 'impedance at the injection point and the amplitude peak corresponding to the reflection of the signal on the end of the cable.

FIG. 3 schematically shows, in a flowchart, the main steps in carrying out a fault detection method according to the invention.

The method is described below by taking a pulse signal as an example of a test signal. However, as indicated above, the invention can be extended to other time or frequency reflectometry signals and also multicarrier signals of the OMTDR or type. MCTDR. In this case, as indicated above, the reflectogram is obtained by inter-correlating the reflected signal R and the injected signal S.

The first step 300 of the method consists in defining several test signals each having a different pass band. For example, for the case of time reflectometry, the test signal chosen is a time pulse, for example a Gaussian pulse. The different test signals are defined 300 by varying the maximum frequency of the basic test signal. The choice of maximum frequencies and the number of frequencies are parameters of the invention. The number of frequencies (and therefore of test signals) increases the complexity of implementing the invention but improves the performance of fault detection. The choice of maximum frequencies also depends on the characteristics of the cable to be analyzed and the type of fault to be detected.

Subsequently, each test signal is set by its maximum frequency. When we talk about the frequency of a signal in the remainder of the text, this is in each case the maximum frequency of the signal.

From the defined test signals, then, in a step 301, initial reflectometry measurements associated with each test signal and corresponding to the cable to be analyzed in an initial state are determined. The initial state of the cable corresponds, for example, to a new cable which is considered faultless, that is to say a healthy cable. The initial state of the cable can also correspond to an intermediate state of the cable after having performed the method according to the invention for the first time and detecting the presence of a fault in the cable. In this second case, the invention aims to monitor the evolution of a defect initially detected and / or the appearance of new defects.

In a first embodiment, the initial reflectometry measurements are obtained by performing a reflectometry test for each test signal. The reflectometry test consists of injecting the test signal into the cable and then measuring the reflection of this signal. In the event that the test signal is a signal more complex than a temporal pulse, the reflectometry test also includes an intercorrelation calculation between the measurement of the reflected signal and a copy of the test signal.

In a second embodiment, the initial reflectometry measurements are obtained by simulating the propagation of the test signals in the cable from known parameters of the cable, for example parameters R, L, C and G. In fact, the signals reflected can be simulated from the equations of telegraphs, especially when the cable is sound.

In a particular embodiment, the reflectometry measurements are temporally truncated by applying a temporal truncation window around the area corresponding to the pulse that is to be analyzed.

At the end of step 301, a multidimensional data matrix X is obtained which contains, in each column, the samples of an initial reflectometry measurement corresponding to a test signal defined for a given maximum frequency. More simply, the number of columns of the matrix corresponds to the different maximum frequencies of each test signal and the number of rows of the matrix corresponds to the number of samples per reflectometry measurement.

From this data matrix X, a model is determined 302 by principal component analysis.

Principal component analysis is a statistical method applied to multidimensional variables. It is, for example, described in the reference documents [8] and [9].

Principal component analysis is based on the use of redundancy in multidimensional data which is correlated with each other. The original data are projected into a subspace of reduced dimensions defined by the so-called principal components among all the dimensions. A general objective of component analysis principal concerns the reduction of the dimensions of the starting space, in other words the search for the principal components. A few steps of this statistical method are detailed below which are applied, in the present case, to reflectometry measurements in order to determine a model of the cable in its initial state.

The original data matrix X contains n observations for m dimensions. It is normalized from the mean p _Xj and the standard deviation a _Xj of the values of each vector X _j . Each sample X _jC of the normalized matrix X _c is, for example, calculated via the following relation (1):

[Math.1]

Xj-Vxj

Xjc =

& Xj (1) We then calculate the covariance or autocorrelation matrix C from the normalized data matrix X _c , with n the number of observations, in other words the number of samples per reflectometry measurement. [Math.2]

C = ^ X _c ^T X _c (2)

By applying a singular value decomposition, the matrix C can be rewritten as follows: [Math. 3]

C = PAP ^T , with PP ^T = P ^T P = I _mxn (3)

P is the matrix of the eigenvectors of the matrix C. The matrix A is a diagonal matrix which contains on its diagonal the eigenvalues {Aj} of the matrix C. The normalized data matrix X _c is then broken down into two parts. The first term corresponds to the principal space and the second term corresponds to the residual space. I is the number of principal components.

[Math. 4]

The matrix Pi contains the first I eigenvectors of the matrix C. These eigenvectors define the main space. The matrix P contains the other (last) eigenvectors of the matrix C which define the residual space.

Equation (4) defines the model of the cable in the initial state from the multidimensional data X _c .

The determination of the number I of principal components is a parameter of the method. It can be chosen, for example, from a criterion of the type cumulative percentage of variance or "Cumulative Percent Variance" in English. This criterion consists in choosing I such that the ratio of the sum of the eigenvalues of the principal components relative to the sum of all the eigenvalues is greater than or equal to a given percentage, for example 90%.

[Math 5]

CPH 0 =! ¾> 9 °% (5)

The model by principal component analysis is determined for an initial state of the cable at an initial time.

From this model, the invention then consists in analyzing the same cable or a similar cable when it has aged in order to compare new measurements. reflectometry to the initial model determined in step 302, with a view to detecting the presence of a defect.

Thus, in a step 303, new reflectometry measurements are carried out, identical to the measurements carried out in step 301, but applied to the cable in a state different from the initial state.

The reflectometry measurements are carried out with the same test signals 300. Then, in a step 304, these new measurements are compared with the model determined in step 302.

The comparison step 304 is, for example, performed by projecting a multidimensional matrix containing the new measurements in the main space or in the residual space, according to the model defined in step 302.

The respective projections of the new matrix of measures X _ne in the principal space T _ne w and in the residual space T _new are, for example, determined using the following relations: [Math. 6]

T _n ew -% new ^ l (6)

[Math. 7]

f _new = x _new - PiP? (7)

Then an error is determined between the new measurements and a prediction of these measurements from the model, in other words a prediction of the nature of these measurements for a cable in a state close to the initial state.

This error can be calculated in the principal space or in the residual space. To calculate the prediction error in residual space, we use a statistical criterion such as the quadratic prediction error or “Squared prediction error” SPE, denoted Q. [Math.8]

Q = T ¹ new T ¹ n ^T ew

To calculate the prediction error in the principal space, one uses a statistical criterion of Hotelling, noted T ² . [Math. 9]

t _j are the column vectors of the matrix T _ne w

A _j are the eigenvalues determined in step 302.

The Q and T ² statistics are described in the publication [9]. They are calculated in the form of vectors comprising a value for each row of the matrix of measurements X _ne w Recall that the rows of the matrix X _ne w correspond to the sampling instants of the reflectometry measurements and the columns correspond to the different test signals associated with different frequency bands.

A prediction error is thus calculated for each row of the matrix X _new composed of a sample of each of the reflectometry measurements. In a step 305, the prediction error is compared, for each multidimensional sample, with a confidence threshold. If the prediction error exceeds the confidence threshold, it is deduced that the associated sample corresponds to a fault on the cable. The confidence thresholds Q _a and T ² _a are determined from the initial model of the cable determined in step 302, for example using the following theoretical relationships which are given in document [10]. [Math. 10]

[Math. 1 1]

[Math. 12]

[Math. 13]

c _a is a critical value of the normal distribution with a confidence level a and F _{¾ n.¾ a} is a critical value of the distribution function of

Fisher-Snedecor.

Without departing from the scope of the invention, other confidence threshold values can be determined taking into account the intended application, the aim being to declare the presence of a fault when the error enters the new reflectometry measurement multidimensional and this measurement projected in the space defined by the model determined in step 302 is significant.

An advantage specific to this method is that the model of the initial state of the cable is established from multidimensional data which take into account counts multiple reflectometry measurements for several different test signals with different bandwidths.

Figures 4 to 10 illustrate the implementation of the invention on a particular example. This example is given for illustrative purposes only and should not be construed to limit the scope of the invention.

FIG. 4a illustrates an example of a fault on the sheath of a cable. The fault is present at a position x _f and at a length L _f and has an angular opening 6 _f which takes the values 45 °, 90 ° or 180 ° depending on the severity of the fault.

By way of example, FIGS. 4, 5 and 6 have shown examples of reflectometry measurements obtained for four pulse signals of maximum frequency respectively equal to 1 GHz, 2 GHz, 3GHz and 4GHz.

Figure 4 shows the four measurements 401, 402,403,404 obtained in the case of a defect in the sheath of a cable with an angular opening of 45 °.

Figure 5 shows the four measurements 501, 502,503,504 obtained in the event of a fault in the sheath of a cable with an angular opening of 90 °.

Figure 6 shows the four measurements 601, 602,603,604 obtained in the case of a defect in the sheath of a cable with an angular opening of 180 °.

We see that the amplitude of the peak corresponding to the fault varies significantly depending on the maximum frequency of the test signal and the severity of the fault.

Depending on the fixed detection threshold, which also depends on the signal-to-noise ratio of the measurements, a fault may not be detected if the maximum frequency chosen is too low. Conversely a maximum frequency too high has drawbacks in terms of signal attenuation but also in terms of implementation of the device.

An advantage of the invention is to use several measurements at different maximum frequencies to improve fault detection but also to be able to then select the most appropriate maximum frequency.

Figure 7 illustrates on a diagram, the values of the quadratic prediction error Q for each multidimensional sample corresponding to the respective measurements of Figures 4,5 and 6. The data are concatenated on the x-axis. The first part corresponds to the measurements in figure 4, the second part corresponds to the measurements in figure 5, the third part corresponds to the measurements in figure 6.

The confidence threshold Q _a is represented in FIG. 7. A defect is declared for the samples which exceed the confidence threshold. Thus, it can be observed that the least severe defect (corresponding to the measurements in FIG. 4) is not detected with the choice which is taken for the confidence threshold. The other two faults are correctly detected. FIG. 8 illustrates results of the same type as those of FIG. 7, but this time with prediction error values calculated from the Hotelling statistic T ² . The confidence threshold T ² _a is also represented.

Without departing from the scope of the invention, it is possible to choose either to calculate the prediction error Q and to base the detection test only on this error, or to calculate the prediction error T ² and to base the detection test only on this error, or again to calculate the two prediction errors and to base the detection test on a weighted combination of the two errors. In an optional additional step 306, for at least one sample for which the prediction error has exceeded the confidence threshold, the contribution of each dimension (that is to say each frequency in the present case) to the mistake.

One possibility for calculating this contribution consists in determining, for each frequency, the ratio between the absolute value of the sample Xi (s) for this frequency and the sum of the absolute values of the sample for all the frequencies Xk (s) with k varying from 1 to n. n is the number of frequencies or test signals or dimensions.

[Math. 14]

FIG. 9 illustrates the respective contributions of the different frequencies for the sample 6508 of FIG. 7, this sample having a prediction error which exceeds the confidence threshold. The frequencies are identified on the abscissa of the diagram of FIG. 9. The 1 GHz frequency is not mentioned because its contribution is too low. The 4 GHz frequency has the highest contribution to the sample for which an error is detected. It is therefore this frequency which is selected.

Figure 10 illustrates similar results for sample 10620 in Figure 8. Again, the frequency 4 GHz is the one with the highest contribution to error.

Thus, the objective of step 306 is to select one or more dimensions, that is to say one or more frequencies in the present case, from among all the frequencies used at the start to carry out the reflectometry tests. One possible embodiment of step 306 is to select the frequency which has the highest contribution to the error. This frequency corresponds to the test signal which is most likely to allow detection of the fault.

In the examples given in Figures 9 and 10, this is the frequency 4 GHZ.

The selected frequency can then be used to perform a conventional OTDR test with a single test signal at that frequency, to monitor the occurrence of a fault on a similar cable or on the same cable.

Another possible embodiment of step 306 is to select one or more frequencies from among those that have the highest contributions to the error (for example, the 4GHz and 3GHz frequencies in the example of Figures 9 and 10). This reduced set of frequencies is then used during a new execution of the process in order to continue monitoring the development of faults on the cable. Thus, the method is performed again by reducing the number of test signals to two signals instead of four, in the example given above.

In general, the number and choice of frequencies can be adapted according to the intended objective. For example, when it comes to continuous monitoring of a cable, the number of frequencies can be reduced from the first execution of the process if it is only to monitor the evolution of a fault detected. On the contrary, if one seeks to monitor the appearance of other faults over time, the initial number of frequencies can be maintained.

Likewise, after step 305, if no fault is detected, the method is iterated with the same test signals with a view to continuously monitoring the condition of the cable. Thus, the invention makes it possible to select the most suitable frequency and thus optimize the choice of the test signal to be used for future reflectometry tests.

The method according to the invention can be implemented as a computer program, the method being applied to reflectometry measurements r acquired using a reflectometry device of the type described in FIG. The invention can be implemented as a computer program comprising instructions for its execution. The computer program may be recorded on a recording medium readable by a processor.

Reference to a computer program which when executed performs any of the functions described above is not limited to an application program running on a single host computer. Rather, the terms computer program and software are used herein in a general sense to refer to any type of computer code (eg, application software, firmware, firmware, or any other form of computer code). computer instruction) which can be used to program one or more processors to implement aspects of the techniques described herein. The computing means or resources can in particular be distributed (“Cloud computing”), possibly according to peer-to-peer technologies. Software code can be run on any suitable processor (e.g., microprocessor) or processor core, or a set of processors, whether provided in a single computing device or distributed among multiple computing devices (e.g. example such as possibly accessible in the environment of the device). The executable code of each program allowing the programmable device to implement the processes according to the invention can be stored, for example, in the hard disk or in read only memory. In general, the program or programs can be loaded into one of the storage means of the device before being executed. The central unit can controlling and directing the execution of the instructions or portions of software code of the program or programs according to the invention, instructions which are stored in the hard disk or in the read-only memory or else in the other aforementioned storage elements.

Alternatively, the invention can also be implemented by means of a processor embedded in a specific test device. The processor can be a generic processor, a specific processor, an integrated circuit specific to an application (also known under the English name of ASIC for “Application-Specific Integrated Circuit”) or an array of programmable gates in situ (also known as the English name of FPGA for “Field-Programmable Gâte Array”). The device according to the invention can use one or more dedicated electronic circuits or a general purpose circuit. The technique of the invention can be carried out on a reprogrammable computing machine (a processor or a microcontroller for example) executing a program comprising a sequence of instructions, or on a dedicated computing machine (for example a set of gates logic such as an FPGA or ASIC, or any other hardware module).

FIG. 11 schematically shows, on a block diagram, an example of a reflectometry device capable of implementing the method according to the invention.

A reflectometry device, or reflectometer, comprises at least one signal generator GS, to generate a test signal s and inject it into the cable to be analyzed CA which has a non-clear defect DNF, Ml measuring equipment to measure the reflected signal r in the AC cable and an electronic component MC of the integrated circuit type, such as a programmable logic circuit, for example of the FPGA type or a micro-controller, for example a digital signal processor, which receives a measurement of the reflected signal r (t) and is configured to execute the method according to the invention in order to detect one or more non-blunt faults. The electronic component MC can also include both an integrated circuit, for example example for carrying out the acquisition of the reflected signal, and a microcontroller for carrying out the processing steps required by the invention.

The injection of the test signal s into the cable is, for example, carried out by a coupling device (not shown in FIG. 11) which can be a capacitive or inductive effect coupler or else using a ohmic connection. The coupling device can be realized by physical connectors which connect the signal generator to the cable or by non-contact means, for example by using a metal cylinder whose internal diameter is substantially equal to the external diameter of the cable and which produces an effect capacitive coupling with the cable.

Acquisition of the signal reflected in the cable can also be achieved by means of a coupling device of the type described above.

The reflectometry system may also include a digital-to-analog converter disposed between the test signal generator, in the case where it is a digital signal, and the injection coupler.

The reflectometry system may also include an analog-to-digital converter disposed between the reflected signal measurement coupler and the M1 measurement equipment or the MC electronic component for the purpose of digitizing the measured analog signal.

In addition, a processing unit (not shown in FIG. 11), of the computer, personal digital assistant or other type is used to control the reflectometry system according to the invention and display the results of the measurements on a man-machine interface. .

The displayed results may include one or more reflectograms calculated using the method according to the invention and / or information relating to the existence and location of a fault on the cable also produced by the method according to the invention. . The displayed results can also comprise one or more frequency bands selected by the invention to be used for the diagnosis of faults on a given cable. According to a particular embodiment, the injected test signal s can also be supplied to the component MC when the processing operations require knowledge of the injected signal, in particular when these include a step of intercorrelation between the test signal s and the reflected signal r.

The injection of the signal into the cable and the measurement of the reflected signal are, for example, carried out by one and the same component but also by two separate components, in particular when the injection point and the measurement point are dissociated.

The device described in Figure 11 is, for example, implemented by an electronic card on which are arranged the various components. The card can be connected to the cable by a coupler.

In addition, a processing unit, such as a computer, personal digital assistant or other electronic or equivalent computer device can be used to control the reflectometry device and display the results of the calculations performed by the component MC on a man-machine interface, in in particular, information on the detection and location of faults on the cable.

References

[1] F. Auzanneau, "Wire troubleshooting and diagnosis: review and perspectives", Progress In Electromagnetics Research B, vol. 49, pp. 253-279, 2013.

[2] C. Furse, Y. C. Chung, C. Lo, P. Pendayala. "A critical comparison of reflectometry methods for location of wiring faults", Journal of Smart Structures and Systems, 2 (1), pp. 25-46, 2006.

[3] S. Sallem, N. Ravot, "Self-adaptive correlation method for soft defect detection in cable by reflectometry", Proceeding of 2014 IEEE Sensors Conference, 2114-2117, Valencia, Spain, 2014.

[4] S. Sallem, N. Ravot, "Self-adaptive correlation method for soft defect detection in cable by reflectometry". In Sensors, 2014 IEEE, pp. 21 14-21 17, 2014.

[5] W. Ben Flassen, M. Gallego Roman, B. Charnier, N. Ravot, A. Dupret, A. Zanchetta, F. Morel, "Embedded OMTDR Sensor for Small Soft Fault Location on Aging Aircraft Wiring Systems", Procedia Engineering, vol. 168, pp. 1698-1701, 2016.

[6] L. A. Griffiths, R. Parakh, C. Furse, B. Baker, "The invisible fray: A critical analysis of the use of reflectometry for fray location". IEEE Sensors Journal, 6 (3), 697-706, 2006.

[7] M. Smail, Y. Le Bihan, L. Pichon, "Fast diagnosis of transmission lines using neural networks and principal component analysis." International Journal of Applied Electromagnetics and Mechanics 39, no. 1-4 (2012): 435-441.

[8] J. Harmouche, C. Delpha, D. Diallo, “Incipient Fault Detection and

Diagnosis Based on Kullback - Leibler Divergence Using Principal

Component Analysis: Part I, ”Signal Processing, Elsevier, Vol. 94, January 2014, pp. 278-287.

[9] J. Harmouche, C. Delpha, D. Diallo, “Incipient Fault Detection and

Diagnosis Based on Kullback - Leibler Divergence Using Principal Component Analysis: Part II, ”Signal Processing, April 2015, Vol. 109, pp. 334-344.

[10] D. Sliskovic, R. Grbic, Z. Hocenski, "Multivariate statistical process monitoring". Tehnicki Vjesnik-Technical Gazette, 19 (1), 33-41, 2012.

Claims

1. A method of detecting non-clear faults in a transmission line (L) comprising the steps of:

- Define (300) several test signals each having a different maximum frequency spectral band, all of the test signals forming a first multidimensional variable,

- Apply (302) to the first multidimensional variable, a principal component analysis to determine a model of the transmission line from the measurements of signals characteristic of the reflection of said test signals previously injected into the line (L) at an instant initial, each measurement being called a temporal reflectogram,

- Acquire (303) several time reflectograms corresponding to said test signals previously injected into line (L) at a current instant so as to form a second multidimensional variable,

- Determine (304) at least one prediction error between the second multidimensional variable and a projection of the second multidimensional variable on a space defined by the model of the transmission line at the initial time,

- Compare (305) the at least one prediction error with a fault detection threshold and deduce therefrom the existence of a fault when at least one prediction error exceeds a fault detection threshold for at least one sample of the multidimensional variable.

2. A method of detecting non-clear defects according to claim 1 wherein the space defined by the model of the transmission line at the instant. initial is a principal subspace or a residual subspace defined by the analysis in principal components.

3. Method for detecting non-blunt defects according to one of the preceding claims, in which the prediction error is taken from an error of the quadratic prediction error type or an error of the Hotelling statistical type T ² .

4. A method of detecting non-frank faults according to one of the preceding claims, in which the initial state of the line is a state for which the line has no fault.

5. Method for detecting non-frank defects according to one of the preceding claims comprising the step of, for each sample of the multidimensional variable for which the prediction error exceeds a detection threshold, determining (306) a relative contribution of each frequency to that sample.

6. A method for detecting non-blunt defects according to claim 5 comprising a step of selecting (306) at least one frequency among the frequencies which contribute the most to the samples for which a defect is detected.

7. A method for detecting non-frank defects according to one of the preceding claims, in which the defect detection threshold is a confidence threshold determined from the model obtained by the principal component analysis.

8. Method for detecting non-frank defects according to one of the preceding claims, in which the test signals are time pulses of variable maximum frequency.

9. A method for detecting non-frank defects according to one of claims 1 to 7 wherein the test signals are multi-carrier signals of variable maximum frequency.

10. A method for detecting non-frank defects according to claim 9 wherein the acquisition (301, 303) of a time reflectogram comprises the correlation of the measurement of the signal characteristic of the reflection of a test signal previously injected into the line (L) and a test signal.

11. Device for the analysis of a cable comprising a measuring device (M1), at a point of the cable (CA), of a signal reflected in the cable and a computer (MC) configured to carry out the detection process. non-clear defects according to one of the preceding claims.

12. A computer program comprising instructions for carrying out the method for detecting non-frank faults in a cable according to any one of claims 1 to 10, when the program is executed by a processor.

13. Recording medium readable by a processor on which is recorded a program comprising instructions for the execution of the method for detecting non-clear faults in a cable according to any one of claims 1 to 10, when the program is executed. by a processor.