WO2012042142A1

WO2012042142A1 - Method of detecting defects of an array by reflectometry and system implementing the method

Info

Publication number: WO2012042142A1
Application number: PCT/FR2011/052038
Authority: WO
Inventors: Adrien Lelong
Original assignee: Commissariat A L'energie Atomique Et Aux Energies Alternatives
Priority date: 2010-09-30
Filing date: 2011-09-07
Publication date: 2012-04-05
Also published as: EP2622360A1; FR2965629A1; FR2965629B1

Abstract

The subject of the invention is a method for detecting defects of an array by reflectometry comprising a step (200) of measuring the response h_cur of the array, a step (201) of detecting defects by analyzing the difference h_Acur between this measurement h_cur and a reference response h_mem. The reference response h_mem is updated using stored measurements h_cur of the response of the array. The subject of the invention is also a reflectrometry system implementing the method.

Description

The present invention relates to a method for detecting faults in a network by reflectometry and a system implementing the method. It applies in particular to the fields of OTDR in electronics and optoelectronics. Cables are ubiquitous in all electrical systems, for powering or transmitting information. These cables are subject to the same constraints as the systems they connect and may be subject to failures. It is therefore necessary to be able to test their status and provide information on the detection of defects, but also their location and type, to help with maintenance and prevention. For this, so-called reflectometry methods are implemented. These can also be used to detect faults in an optical fiber network.

The principle of OTDR is based on the injection of a test signal. The shape of this signal changes significantly during their propagation back and forth in a cable, these changes being the consequence of the physical phenomena of attenuation and dispersion.

The OTDR methods use a principle similar to that of the radar: an electrical signal, the test signal, often of high frequency or wide band, is injected in one or more places of the cable to be tested. Said signal propagates in the cable or network and returns a portion of its energy when it encounters an electrical discontinuity. An electrical discontinuity may result, for example, from a connection, the end of the cable or a fault. The analysis of the signals returned to the injection point makes it possible to deduce information on the presence and the location of these discontinuities, thus possible defects. An analysis in the time domain is usually performed. Several temporal methods exist, such as the Time Domain Reflectometry (TDR) method, the Time Domain Reflectometry (STDR) method, the TDR Spread Spectrum (SSTDR) method, and the MultiCarrier TDR (MCTDR) method. The test signal depends on the OTDR method used. The signal resulting from the reflectometry is called reflectogram and consists of a plurality of peaks corresponding to the singularities of the network. There may be several peaks by singularity, some corresponding to multiple reflections. In a complex network, the objective is then to determine which peak corresponds to a fault and then isolate it correctly in order to precisely locate said fault.

Frequency methods can also be used in the context of OTDR. However, when the duration of the measurement is an important criterion, temporal methods are preferable.

For the detection of a fault, the time approaches commonly adopted, for example in an on-board system, rely on the detection of the largest peak as described in patent US6868357 or on the detection of the first peak exceeding a given threshold. as described in international application WO / 2010/032040.

In order to correctly detect a fault within an arbitrarily complex network, it is necessary to measure the variations of the electrical response of said network. The difficulty is then to identify the first peak that corresponds to the defect in order to deduce its position.

Thus, the comparison of the successive measurements of this electrical response can make it possible to detect the appearance of a fault but also to locate said fault effectively by calculating the difference of the measured responses with and without the fault. However, for reliable detection, the analysis of difference signals h ^ -h ^ between two successive responses

two moments k-1 and k is not enough because often too random. This approach, described in US2009 / 0228223, is problematic, particularly when the defect is slowly establishing itself. Throughout the description, an x-type notation designates a vector, i.e. a set of values organized in a matrix comprising a single column.

Furthermore, in the case where there is a clear appearance of the defect, we have the answers before and after the defect but we can only have one occurrence of this difference. In particular, it is impossible to improve the signal-to-noise ratio by using several difference signals. It is also possible to compare each measurement with a fixed reference as is the case in US6937944. However, in addition to the fact that this requires a calibration phase of the diagnostic system, this solution lacks robustness in that the normal state of the network tested may vary over time.

An object of the invention is in particular to overcome the aforementioned drawbacks.

For this purpose, the subject of the invention is a method for detecting defects of a network by reflectometry comprising a step of measuring the response _CMr of the network, a step of detecting defects by analyzing the difference h _Acur between this measurement. h _∞r and a reference response h _mem . The reference h _mem response is updated using measurements h _oer stored in the network response.

According to one aspect of the invention, a network fault is detected by determining whether the standard of the difference signal h _Acur is less than a predefined threshold T.

According to another aspect of the invention, the standard of the difference signal h _Acur is of standard type L1 or L2 standard or L∞ standard.

In one implementation mode, the reference response h _mem is updated when no fault is detected.

The invention provides that the reference response h _mem is updated when a change of state of the network of a duration greater than a predefined value is detected.

The reference h _mem response can be updated using the following expression: ^{^)} & where

the exponent (k) designates the current measurement instant and the exponent (k-1) the instant preceding the instant (k);

a represents the forgetting factor of the low-pass filter.

The threshold T is set, for example, using the following expression:

in which :

N is the number of samples of the signal;

c ² is the average variance of the noise;

β is a margin factor.

The invention advantageously provides that when a fault is detected, the difference signal h _Acur is deconvolved, the result of the deconvolution being used to locate and characterize the defects.

According to one aspect of the invention, the method comprises a step of detecting multiple occurrences belonging to the same defect, an average of the difference signals h _Acur being determined when several values of h _Acur obtained successively correspond to the same defect, said mean being stored for use in locating and characterizing said defect.

Two difference signals A _: and A ₂ are considered to belong to the same defect when, for example, an estimator J _Lx is less than a predefined threshold T _J} the estimator being determined using the following expression:

The invention also relates to a reflectometry system comprising means for measuring the response of a network, means for comparing this measured response with a reference response, means for detecting and locating faults appearing in the network. The system implements the method described above.

The network is, for example, an electrical cable network or an optical fiber network. Other features and advantages of the invention will become apparent with the aid of the following description given by way of non-limiting illustration, with reference to the appended drawings in which: FIG. 1 illustrates the principle of a time domain reflectometry system for the diagnosis of in-line cables;

FIG. 2 gives an example of a simplified diagram of the three main stages of a wired network diagnostic method by reflectometry;

FIGS. 3A and 3B give examples of reflectograms obtained for an example of a network with and without defects;

FIG. 4 gives a first example of implementation of the differential detection method according to the invention;

FIG. 5 gives a second example of implementation of the differential detection method according to the invention;

FIG. 6 gives an example of a reflectogram obtained after application of the method according to the invention.

Figure 1 illustrates the principle of a time domain reflectometry system for the diagnosis of in-line cables. The reflectometric analysis is performed online when it is performed on a system while it is running.

The system comprises a test signal generator 100. The test signal is usually generated in digital form and then introduced into the test cable network 102 after being converted to an analog signal x (t) using a digital-to-analog converter CNA 101. The analog signal y (t) resulting in particular from the reflection of x (t) by the network 102 is then converted into a digital signal by a CAN-to-digital converter 103. The converters 101, 103 are connected to the cable network 102 via for example, a coupling system with or without galvanic isolation.

An electrical response measurement is generally affected by a not inconsiderable noise adding to the useful signal. Thus, an averaging operation 105 is applied after the analog-to-digital conversion to improve the signal-to-noise ratio. For example, M measures of electrical signals are realized then their results are accumulated in an average vector noted y.

The time T _mes required to perform a complete measurement can be determined using the following expression:

T _mes = MNT _S , (1) in which

T _s represents the sampling period of the test signal x (t);

N is the number of samples of the signal.

In order to have a lower technological constraint on the digital analog converter 103, it is possible to use equivalent time sampling methods as described in the international application WO / 2009/087045, in order to reach a frequency of sampling of the signal measured higher than that permitted by the converter. In this case, the duration of the measurement is given by the expression:

T _mes = MK'NT _s (2) in which K 'represents the equivalent time factor.

The measurement time is particularly important for the detection of network faults appear intermittently because the duration of these faults can be very low. In the following description, this type of fault is called intermittent fault. In the presence of such defects, the number of measurements M remains limited in order to guarantee an acceptable capture speed. Therefore, it may be necessary to increase the signal-to-noise ratio upstream with other treatments.

This is based on the received signal y (t) after digital to analog conversion the reflectometry system will estimate the response _heart representative of network conditions using, for example, a matched filter.

The measured response can also be chosen, for example, such that h _cur =, this choice depending on the properties of the noise and the test signal.

We then obtain a series of estimates of the rated channel response where k is the time at which each estimate is captured. Differential detection 106 for detecting defects is then performed. The principle is to compare successive measurements of the electrical response with, for example, a reference response. When a fault is detected, the difference signal between a reference response corresponding to the response of the network considered as flawless and that measured with the fault is transmitted to a functional module performing a deconvolution of the signal.

The difference signal is processed by deconvolution 107 to separate the different peaks that compose it. Different pulse deconvolution algorithms can be used for this.

A location processing 108 then aims to determine the distance separating the reflectometer from the fault from the position of the first peak of the result obtained. Figure 2 gives an example of a simplified diagram of the three main steps of a cable diagnosis method by reflectometry. A first step 200 corresponds to the acquisition of the measurements of the response of the tested network, a second step corresponds to the detection of the defects 201 of the network tested and a third step, performed when there is detection, corresponds to the location 202 of these defects in said network.

FIGS. 3A and 3B give examples of reflectograms obtained for an example of a network with and without faults.

These reflectograms correspond to reflectometry results obtained on a lattice with and without defects, before 305, 307 and after 306, 308 high resolution processing. The network 300 considered in this example comprises four line sections. A first section of length l ₀ corresponding to a 50Ω coaxial line has one of its ends used as input of the network. Its other end is connected directly to the first end of a second section 302 twisted pair type and length //. The second end of the second section 302 corresponds to a junction between three sections, the latter being connected to the first end of two sections 303, 304 of twisted pair type and of respective lengths h and h, the other ends of these two sections being left in open circuit. Figure 3A shows the case where there is no fault. A negative peak 309 appears at the junction followed by positive peaks 310, 31 1, 312, 313 corresponding to the ends of lines 303, 304 of lengths h and

FIG. 3B shows curves 307, 308 representative of the state of the network in the presence of a fault. In this example, the fault corresponds to a short circuit appearing at the junction. When the short-circuit occurs, the negative peak 314 corresponding to the junction becomes larger and the peaks 315, 316, 317 correspond to the secondary echoes of this short-circuit. The ends of lines are no longer visible.

Automatic processing to interpret these curves to deduce the presence of a defect is in practice difficult to implement. It is possible to use topology reconstruction algorithms as disclosed in US7282922. However, it is difficult to implement this type of solution when real-time and on-board detection is required.

FIG. 4 gives a first example of implementation of the differential detection method according to the invention. The principle of the method is to use for the detection of defects a reference stable state h _mem taking into account the changes in the state of the network over time. The responses h ^! - measured at each instant k are then compared to this reference.

Thus, initially, the latest measure _heart is acquired 400. This measurement is compared to the steady state reference. This comparison is made initially by determining the standard of the difference signal _Acur corresponding to the difference between _^ and the stable state of reference _mem . The norm of this difference is then compared 401 to a threshold T. This comparison corresponds to the inequality: l _.Acur I <T with | h _AcMr || = | h _CMr - _mem || (3)

As recalled earlier in the description, the presence of a defect is manifested in particular by a peak in the reflectogram. The most natural choice for determining the standard is thus to use the maximum or the infinite standard known to those skilled in the art. In the absence of change, the maximum of the signal corresponds to the amplitude of the noise, when this value is greater than the maximum amplitude of the noise a variation is detected.

It is also possible to use the standard called L2 norm and defined by the following expression: lli = Σ ^ ( ⁴ )

In which the sign || ^e || ₂ refers to the L2 standard.

Concerning the threshold T, this one makes it possible to distinguish the cases where there is no variation of the cases where there is variation compared to the reference answer. In the case where there is no variation, h _Acur contains only noise.

The value of the threshold T used in expression (3) should be judiciously chosen. For example, the threshold T can be set using the expression:

Τ = βΝσ ² , (5) in which

N is the number of samples of the signal;

c ² is the average variance of the noise;

β is a factor corresponding to a margin, the latter being chosen so that the energy of the noise never exceeds the threshold T; the value of β depends on the shape of the noise over time, so for a noise of a pulse nature a higher value will be chosen.

Alternatively, the standard of the difference signal h _{& cur} can be determined on the basis of the so-called standard L1 defined by the expression:

n in which the || · || refers to the L1 standard. The use of the L1 standard is interesting in some cases in terms of performance and computational complexity. It is particularly preferred in the presence of a noise of impulsive nature. In other words, the choice of the standard is optimized according to the nature of the noise.

The standard L2 is taken as an example in the remainder of the description without this choice limiting the scope of the invention. The generic symbol || · || to refer to any chosen standard.

In practice, the value of σ is rarely known in advance, so it is advantageous to be able to estimate it automatically. For this, the variance of the instantaneous difference can be used, which can be determined by using the following expression: î hLMnst

It is possible to demonstrate that the variance of the noise in h _Ainst is equal to 2σ ² . An average estimate of the variance noted at ² can then be calculated using a first-order filter.

At time k, the estimated variance at ² can be determined using the expression:

(k) _. (tl) » ²

σΜ = "- ^G (ki) + (1 -" _σ ) - (8)

2N

In which :

α _σ is the forgetting factor of the filter.

The estimate can then be used directly to calculate the value of the threshold T in real time. When a standard other than the L2 standard is used, the expression (8) can be used by replacing the L2 standard with another standard.

In the case where the difference h _{& heart} equals or exceeds 401 the threshold T, that is to say where there is a significant change in network status, a count time counter 407 is compared to a threshold count _max . The threshold count _max corresponds, for example, to the maximum duration of a fault for it to be considered transitory or intermittent. If count> count _max , the h _mem must be updated. In the case where count <count _max , the count counter is increments 408.

The 407 test makes it possible to take into account the permanent state changes of the network and not to treat them in the same way as a default appearing and disappearing. In the example of Figure 4, it avoids being stuck at the detection of a change of state.

A step 410 then aims to select the record h _Acur stored with the maximum energy level. This selection makes it possible to avoid keeping records corresponding to a transient state of the network. Thus, if> then the value stored in the difference h _Amem is h _Amem = _Acur . The h _Amem signal is the final result of the detection which will be analyzed later for the location of the defects.

On the other hand, keeping the largest difference, the signal-to-noise ratio of the _amem signal is improved.

401 in the case where the difference h _cur is less than the threshold T, that is to say in the case where there is no significant change in the network status, the counter count is set to zero 402.

A test 403 then checks if || h _mem || > 0. If this is the case, it is because there is a fault and the signal h _Amem is analyzed 404 for example by deconvolution to be able to precisely locate faults in the network. h _Amem is then set to zero 405.

Whatever the value of || h _mem || , the h _fflr measurement is used to update the reference state h _mem 406 using a low-pass filter in order to avoid taking into account too fast changes in the state of the network. This low-pass filtering can be implemented using for example the expression:

in which

the exponent (k) designates the current moment; a represents the forgetting factor of the low-pass filter, this factor being directly related to the time constant by a = e ^τ where τ is the time constant of the filter expressed in seconds.

Advantageously, the averaging effect induced by this filtering leads to a quasi-zero noise on the reference signal h _mem . It can be shown that in the case of a white noise, the signal-to-noise ratio of the reference is given by the expression:

SNR _h = - ^l ^ SNR _h , (10) in which

SNR _h is the signal-to-noise ratio of the acquired signals.

Advantageously, the method may be implemented in a distributed reflectometry system such as that described in the article by A. Lelong, L. Sommervogel, N. Ravot, and MO Carrion titled Distributed Reflectometry Method for Wire-fault Leasing Using Selective Average , Sensors Journal, IEEE, pages 300-310, February 2010.

FIG. 5 gives a second example of implementation of the differential detection method according to the invention.

Deconvolution methods used for _Amem difference _processing usually require a high level of signal-to-noise ratio. In a disturbed environment, it may happen that the signal-to-noise ratio of the result of a detection is low. But it is common for the same intermittent fault to occur several times in a row. It is then possible to use the multiple occurrences of the same defect to average the difference signals and thus improve the signal-to-noise ratio obtained.

However, before averaging multiple detections, it is helpful to ensure that these occurrences match the same intermittent fault. In practice, it is not possible to guarantee that an intermittent fault produces an echo of the same amplitude from one occurrence to another, but the position of the peak corresponding to the fault is always the same. So, to check if the results of two detections for two signals from difference Δ ^ and Δ ₂ correspond to the same defect, it is possible to test the value of an estimator J, which value can be obtained by using the expression: J (A ₁ , A ₂ ) = min || A ₁ - λΔ ₂ || (1 1) in which:

λ is a real coeficient whose function is to compensate for any difference in amplitude that may exist between the two occurrences.

The test of the value of J amounts to testing the collinearity of the vectors i and Δ ₂ . If the L2 standard is used, the value of this minimum can be calculated directly using the expression:

When another standard is used, we can use the more generic result of the following expression:

It is then possible to use this relation to compare the two difference signals Δ _: and Δ ₂ . An alternative of the exemplary implementation of the method of FIG. 4 is obtained, this alternative being presented in FIG. 5.

It should be noted that we speak of similarity between two signals when the value of J as defined by the expression (13) is considered low. Thus, it is possible in practice to compare the value J (h _Acur , h _Amem ) with a threshold T which can be parameterized in the same way as for the threshold T described above. This makes it possible to determine if h _Acur is similar to h _Amem 500. Instead of keeping only the maximum of the _amem difference _signals obtained as in the example of FIG. 4, they are all used in order to allow the determination of an average of several values 501 and thus improve the value of the signal-to-noise ratio. Otherwise, the defect is considered new, and the average is reset 502. When several occurrences of h _Acur have been accumulated in h _Amem , the signal obtained can be analyzed by deconvolution. It is then possible to determine the location of the intermittent fault.

The example given here resumes the case presented with Figure 1. The dashed line curve 600 gives the difference between the responses with and without defects. After a high resolution treatment that can be performed by pulse deconvolution, the high resolution curve 601 can be obtained. The number of samples n separating the first peak of the origin makes it possible to calculate the distance between the position of the fault in the cable network and the reflectometer. This distance can be determined using the following expression:

(14) wherein:

T _s is the sampling period of the acquired signal and v is the speed of propagation in the network.

Claims

1 - Method for detecting faults in a network by reflectometry comprising a step (200) of measuring the h _CMr response of the network, a step (201) for detecting faults by analyzing the difference h _Acur between this measurement h _CMr and a reference response _mem , characterized in that the reference response h _mem is updated using stored h _eal measurements of the network response.

Method according to Claim 1, characterized in that a fault of the network is detected (401) by determining whether the standard of the difference signal h _Acur is lower than a predefined threshold T.

Process according to either of Claims 1 and 2, characterized in that the standard of the difference signal h _Acur is of standard type L1 or L2 standard or L∞ standard.

Method according to any one of the preceding claims, characterized in that the reference response h _mem is updated when no fault is detected.

Method according to any one of the preceding claims, characterized in that the reference response h _mem is updated when a change of state of the network of a duration greater than a predefined value is detected.

6. Method according to any one of the preceding claims, characterized in that the reference response h _mem is updated by using the following expression: e _m = ( ¹ - a) h _cur + h _mem , in which

the exponent (k) designates the current measurement instant and the exponent (k-1) the instant preceding the instant (k); a represents the forgetting factor of the low-pass filter.

7- Method according to any one of the preceding claims, characterized in that the threshold T is set using the following expression:

in which

N is the number of samples of the signal;

c ² is the average variance of the noise;

β is a margin factor.

8. Method according to any one of the preceding claims, characterized in that when a fault is detected, the difference signal h _Acur is deconvolved, the result of the deconvolution being used to locate and characterize the defects.

9- Method according to any one of the preceding claims, characterized in that it comprises a step of detecting multiple occurrences belonging to the same defect (500), an average (501) of the difference signals h _Acur being determined when several h _Acur values obtained successively correspond to the same defect, said average being stored for use in locating and characterizing said defect.

10- Method according to claim 9 characterized in that two difference signals Δ _: and A ₂ are considered to belong to the same defect when an estimator J _Lx is less than a predefined threshold 7 ^, the estimator being determined using following expression:

1 1 - OTDR system comprising means for measuring the response of a network, means for comparing this measured response with a reference response, means for detecting and locating the defects appearing in the network characterized in that it sets implement the method according to any one of the preceding claims.

12- System according to claim 1 1 characterized in that the network is a network of electric cables.

13- System according to claim 1 1 characterized in that the network is an optical fiber network.