method and system for non-destructive testing of an elongated structure such as a rail

Abstract

Non-destructive testing method comprising: acquisition (104) of signals () representative of ultrasonic waves circulating in a structure; determination (106) of a range of values ​​() of a physical quantity reached during the first phase; estimation (108) of a model () representative of a reference state of the structure; calculation (110) of a threshold (); acquisition (202) of a new signal (); measurement (204) of a new value () of the physical quantity during the acquisition of the new signal (); from the new signal () and the model (), determination (208) of a degradation index of the structure (); updating the model () using the new signal () if: the new value is outside the range of values ​​() and the degradation index of the structure () has not crossed the threshold (); updating the range of values ​​to include the new value. Figure for the summary: Figure 2

Description

Title of the invention: method and system for non-destructive testing of an elongated structure such as a rail Technical field

[0001] This document relates to a method and system for non-destructive testing of an elongated structure such as a rail. TECHNOLOGICAL BACKGROUND

[0002] Methods for non-destructive testing of a structure are known.

[0003] Some of these methods estimate the health status of a structure based on signals emanating from the structure, and therefore carrying information about this structure. These signals are acquired using sensors.

[0004] A problem that arises with such methods is that of false alarms. A false alarm occurs when the structure is diagnosed as having a defect, when in reality such a defect is not present. Statement of the invention

[0005] An object of the invention is to control an elongated structure in a non-destructive manner, in a manner which strives to reduce the occurrence of false alarms.

[0006] To this end, according to a first aspect, a method for non-destructive testing of an elongated structure such as a rail is proposed, the method comprising: • during a first phase, acquisition of reference signals representative of ultrasonic waves circulating in the elongated structure, the acquisition being implemented by an acquisition system sensitive to a physical quantity, • determination of a range of values ​​of the physical quantity reached during the first phase, • from the reference signals, estimation of a reference model representative of a reference state of the elongated structure in which the elongated structure was found during the first phase, • from the reference signals, calculation of a threshold, • during a second phase subsequent to the first phase, acquisition of a new signal representative of an ultrasonic wave circulating in the elongated structure, the new signal being acquired while the elongated structure is in a second state, • measurement of a new value of the physical quantity during the acquisition of the new signal, • from the new signal and the reference model, determination of a

[0007]

[0008]

[0009]

[0010] degradation index of the elongated structure associated with the new signal, the degradation index of the elongated structure associated with the new signal indicating a level of degradation of the second state of the elongated structure compared to the reference state of the elongated structure, • update of the reference model using the new signal if at least the following conditions are met: • the new value of the physical quantity is outside the range of values, and • the degradation index of the elongated structure associated with the new signal did not cross the threshold, • update the range of values ​​so that the range of values ​​includes the new value of the physical quantity. The method according to the first aspect may further comprise the following optional features, taken alone or in combination where technically possible. Preferably: • the acquisition step during the second phase is repeated so that a sequence of signals starting with the new signal is acquired during the second phase, the sequence comprising h consecutive signals, • the steps of measuring and determining a degradation index are repeated for each of the h consecutive signals, • the update of the reference model is implemented in one go using the h consecutive signals if at least the following conditions are met: • during the respective acquisitions of the h consecutive signals of the sequence, h values ​​of the physical quantity were measured which are all outside the range of values, and • from the h consecutive signals of the sequence and the reference model, h degradation indices were determined, respectively associated with the h consecutive signals which did not cross the threshold. Preferably, the update of the reference model is triggered after the measurement of a value of the physical quantity which enters for the first time into the range of values ​​since the acquisition of the new signal. Preferably, the method according to the first aspect comprises generating a result indicating the presence of a defect in the elongated structure provided that the degradation index of the elongated structure associated with the new signal has crossed the threshold. Preferably: : • the acquisition step during the second phase is repeated so that a sequence of signals starting with the new signal is acquired during the second phase, the sequence of signals comprising m consecutive signals, • the generation of the result indicating the presence of a defect in the elongated structure is implemented if the following conditions are met: from the >n consecutive signals of the sequence and the reference model, h degradation indices respectively associated with the h consecutive signals which have all crossed the threshold have been determined.

[0012] Preferably: • the acquisition step during the second phase is repeated during the second phase, the acquisition system using an acquisition period separating the acquisition times of two consecutive signals, • the measurement step is repeated for each of the signals in the sequence, • the acquisition period used by the acquisition system is adjusted to: • a first duration when at least the following conditions are met: • the new value of the physical quantity is in the range of values, and • the degradation index of the elongated structure associated with the new signal did not cross the threshold, • a second duration strictly less than the first duration, when the new value of the physical quantity is outside the range of values, • a third duration less than or equal to the second duration, when at least the following conditions are met: • the new value of the physical quantity is within the range of values, and • the degradation index of the elongated structure associated with the new signal has crossed the threshold.

[0013] Preferably, the method according to the first aspect further comprises the following steps: • calculation of a variance of the new signal, • generation of a result indicating rail breakage provided that the variance exceeds a predefined variance threshold.

[0014] Preferably, the estimation of the reference model comprises a singular value decomposition of a matrix C formed from the reference signals, the decomposition being carried out as follows:

[0016] where: • S is a matrix comprising the singular values ​​of the matrix C, • U is a matrix comprising left singular vectors respectively associated with singular values, • V is a matrix comprising right-hand singular vectors respectively associated with the singular values,

[0017] the reference model being formed by a subset of vectors selected from among the singular vectors on the left or from among the singular vectors on the right.

[0018] Preferably, the threshold is calculated by the following formula:

[0019] V = gX^h

[0020] where: * h is a chi-square distribution with h degrees of freedom and a level of meaning has, 8 = 9, / 6,

[0021] h = 07 / ¾ • n _ VN Xi, with = 1, 2 • N is the number of acquired reference signals, • is a singular value of the matrix S of index j associated with vectors singulars not selected to form the reference model.

[0022] Preferably, the physical quantity is or depends on: a temperature reached by the elongated structure, a humidity level and / or a pressure experienced by the elongated structure.

[0023] Also provided, according to a second aspect, is a system configured for implementing the method according to the first aspect. The system according to the second aspect is a system for non-destructive testing of an elongated structure such as a rail, the system comprising: • a signal acquisition system sensitive to a physical quantity, the acquisition system being configured to: • during a first phase, acquire reference signals representative of ultrasonic waves circulating in the elongated structure, • during a second phase subsequent to the first phase, acquiring a new signal representative of an ultrasonic wave circulating in the elongated structure, the new signal being acquired while the elongated structure is in a second state, • a measuring system configured to: • determine a range of values ​​of the physical quantity reached during the first phase, • measure a new value of the physical quantity when acquiring the new signal, • a processing module configured to: • from the reference signals, estimate a reference model representative of a reference state of the elongated structure in which the elongated structure was found during the first phase, • from the first signals, calculate a threshold, • from the new signal and the reference model, determine a degradation index of the elongated structure associated with the new signal, the degradation index of the elongated structure associated with the new signal indicating a level of degradation of the second state of the elongated structure compared to the reference state of the elongated structure, • update the reference model using the new signal if at least the following conditions are met: • the new value of the physical quantity is outside the range of values, and • the degradation index of the elongated structure associated with the new signal did not cross the threshold • update the range of values ​​so that the range of values ​​includes the new value of the physical quantity. DESCRIPTION OF FIGURES

[0024] Other characteristics, aims and advantages of the invention will emerge from the following description, which is purely illustrative and non-limiting, and which must be read in conjunction with the appended drawings in which:

[0025] [Fig.l] schematically illustrates a non-destructive testing system for an elongated structure, according to one embodiment.

[0026] [Fig.2] is a flowchart of a step of a method for non-destructive testing of an elongated structure, according to one embodiment.

[0027] [Fig.3], [Fig.4] and [Fig.5] detail sub-steps of the method of [Fig.2].

[0028] [Fig.6] superimposes data curves obtained during the implementation of a witness process.

[0029] [Fig.7] superimposes data curves of the same type as [Fig.6], obtained during the implementation of a non-destructive testing method according to one embodiment.

[0030] Throughout the figures, similar elements bear identical references. DETAILED DESCRIPTION OF THE INVENTION 1) Non-destructive testing system

[0031] With reference to [Fig.l], a non-destructive testing system 1 of an elongated structure R comprises an acquisition system 2, a measurement system 4 and a processing module 6.

[0032] The elongated structure R is for example a rail. It is however understood that this application is not limiting, therefore the system 1 can be used to control other types of elongated structures than rails.

[0033] The acquisition system 2 comprises at least one ultrasonic wave receiver.

[0034] The or each ultrasonic wave receiver is arranged to acquire ultrasonic waves circulating in the elongated structure R. For this purpose, the or each receiver can be fixed on the elongated structure R.

[0035] When several receivers are present in the acquisition system 2, these receivers can be distributed along the elongated structure R. For example, it can be provided that the distance between two neighboring receivers is constant.

[0036] The or each receiver may in particular be configured to convert an ultrasonic wave that it receives (which can be seen as a mechanical vibration) into an electrical signal, in particular a sampled digital signal. The or each receiver may for example be of the piezoelectric type.

[0037] The acquisition system 2 may also comprise at least one ultrasonic wave transmitter.

[0038] The or each transmitter is adapted to generate and transmit ultrasonic waves in the elongated structure R, such that these ultrasonic waves can circulate in the elongated structure R and then be acquired by at least one receiver of the acquisition system 2.

[0039] In particular, the acquisition system 2 may comprise probes which provide both the functions of transmitter and receiver of ultrasonic waves. In other words, the or each receiver may in reality be a transmitter and receiver probe (also referred to as a transmitter-receiver or translator).

[0040] As indicated previously, the elongated structure R may be a rail forming part of a railway track. This railway track comprises two rails. In this case, the acquisition system 2 may comprise at least one receiver adapted to acquire signals representative of ultrasonic waves circulating in one of the two rails, and at least one receiver adapted to acquire signals representative of ultrasonic waves circulating in the other rail; the same applies to any transmitters present in the acquisition system 2.

[0041] The acquisition system 2 is capable of acquiring signals representative of ultrasonic waves sequentially, using an acquisition period. The acquisition period is the duration which separates the respective emission times of two waves consecutively emitted by the acquisition system 2, and which further separates the respective instants of acquisition of the signals representative of these consecutive waves (the acquisition system 2 assumes that the propagation time of the waves is constant). Thus, as long as the acquisition period is constant, the emission of the ultrasonic waves emitted by the acquisition system 2 is periodic, and the acquisition of the corresponding signals is also periodic.

[0042] The acquisition period used by the acquisition system 2 is adjustable to different values, including a value P and a value P' lower than the value P. In other words, signals representative of ultrasonic waves can be sequentially acquired by the acquisition system 2 at a relatively slow rate when the acquisition period has the value P, and signals representative of ultrasonic waves can be sequentially acquired by the acquisition system 2 at a slower rate when the acquisition period has the value P'.

[0043] Furthermore, the measurement system 4 is configured to measure a physical quantity to which the acquisition system is sensitive. By “sensitive”, it is meant that certain excessive variations in the physical quantity are likely to disrupt the operation of the acquisition system, to the point that certain signals acquired by the acquisition system are altered.

[0044] In the following, an embodiment is detailed in which the physical quantity measured by the measuring system is a temperature reached by the elongated structure R, it being understood that other physical quantities can be measured. The measuring system 4 may comprise, in this embodiment, a temperature sensor, or several temperature sensors 4 arranged in different positions along this elongated structure R.

[0045] Each temperature sensor of the temperature measuring system is fixed on the elongated structure R.

[0046] Furthermore, the processing module 6 is configured to carry out processing operations based on signals acquired by the acquisition system 2, and based on temperatures measured by the measurement system 4. These processing operations will be detailed later.

[0047] The processing module 6 comprises or is coupled to a memory 8 adapted to store data accessible by the processing module 6. This data can be used by the processing module 6 or else be produced by this processing module 6.

[0048] The memory 8 is of any type. It may comprise a random access memory (for example of the RAM type) and a read-only memory (for example of the Flash, EEPROM, hard disk, SSD type).

[0049] The processing module 6 typically comprises a processor adapted to execute a computer program, this computer program comprising ins code instructions adapted for implementing the processing mentioned above when the processor executes the program. The program is stored in memory 8.

[0050] The processing module 6 is of any type, programmable or non-programmable (ASIC, FPGA, etc.).

[0051] The memory 8 can be located in any location. This memory 8 can be part of the acquisition system 2, of the processing module 6, or be located in a server remote from the acquisition system 2 and from the processing module 6. This server can be located on site, that is to say close to the elongated structure R, or be remote. In the latter case, the communication of data from the acquisition system 2 and / or from the or each sensor 4 to the memory 8 can be carried out by wire or by wireless radio. 2) Non-destructive testing process

[0052] A method for non-destructive testing of the elongated structure R according to one embodiment will now be described, with reference to the flowchart of [Fig.2].

[0053] This method uses different predefined values, which are for example stored in the memory 8. These predefined values ​​include: • N: a predefined integer, with N > 1 • M ; a predefined integer, with M > 1 • P, P', P": values ​​for the acquisition period (p < P' < p)

[0054] For example, the following values ​​of acquisition period are chosen: P = 1 month, P'= 1 week, P”= 30 seconds.

[0055] The method also uses different variables, for example allocated in the memory 8. These predefined values ​​are initialized in a step 100 as follows: • h = 0 (temperature range exit counter) • m = 0 (threshold exceedance counter) • Acquisition period = P

[0056] The method comprises two successive phases: a first phase, called the learning phase, and a second phase subsequent to the first phase, called the monitoring phase. 2.1) Learning phase

[0057] It is assumed that the elongated structure R is in a certain state during the learning phase, a state which is conventionally called the “reference state” of the elongated structure R in the following.

[0058] Although this is not mandatory, the reference state is preferably a healthy state, or in any case a state considered acceptable. For example, the learning phase precedes the commissioning of the elongated structure R. In this case, the elongated structure R has not yet been used intensively, so its condition can reasonably be considered healthy.

[0059] The following steps are implemented during the learning phase.

[0060] In a step 102, the acquisition system 2 generates ultrasonic waves, such that these ultrasonic waves propagate in the elongated structure R.

[0061] During this step 102, the acquisition system 2 generates an electrical signal (which is characterized by different parameters such as the frequency, the number of cycles, the envelope shape, etc.). This electrical signal is transmitted, for example via coaxial cables, to one (or more) emitters to be excited (at least one). The emitter translates the electrical energy into mechanical energy (in reality, the translator makes a back-and-forth micro-movement causing a local disturbance, which will propagate in the elongated structure R in the form of an ultrasonic wave. Propagating in the rail, the ultrasonic wave (let us call it incident) interacts with the possible defects that may appear in the elongated structure R (crack, corrosion, injury, breakage, etc.).

[0062] N ultrasonic waves are emitted sequentially during step 102, i.e. at different respective times.

[0063] The N ultrasonic waves are emitted periodically, according to the acquisition period.

[0064] The interaction between the incident wave and a defect generates two waves, one of which is reflected (i.e. part of the incident wave turns around) and the other transmitted (i.e. the other part continues its path along the elongated structure R).

[0065] In an acquisition step 104, the acquisition system 2 acquires N signals acquisition of signals A i' • • • ' Av representative of the N ultrasonic waves having circulated in the elongated structure R.

[0066] Two cases are possible: • The reflected wave is received by the translator that initially emitted the incident wave. In this case, the acquisition mode is called “reflection mode” (known more by its English name “pulse-echo”). • The transmitted wave is received by a receiver separate from the transmitter that emitted the incident wave. In this case, the acquisition mode is called transmission mode (known more by its English name “pitch-catch”).

[0067] The N signals • • • ' Av are acquired periodically, according to the acquisition period.

[0068] The N signals -A' • • • ' xn are then transmitted to memory 8, which stores them.

[0069] Each signal x' is a time signal comprising n samples.

[0070] In a measurement step 106, the temperature measurement system 4 measures N temperatures T b TN reached by the rail.

[0071] Each of the temperatures T{ is associated with one of the signals xt. This association is temporal, in the sense that the temperature Tj is measured during the acquisition of the signal xi, or at least very shortly before or after this acquisition. In any case, the time difference between the instant of measurement of the temperature T t and the instant of acquisition of the associated signal xi is negligible compared to the acquisition period used by the acquisition system 2.

[0072] Ultimately, the acquisition steps 104 and measurement steps 106 are generally concomitant.

[0073] The N temperatures T b TN are then transmitted to the memory 8, which stores them.

[0074] The N temperatures T TN form a temperature range [ Tmin, r^] . Tmin is the minimum temperature among the N measured temperatures, and Tjnax is the maximum temperature among the N measured temperatures.

[0075] Subsequently, the processing module implements the following steps on the basis of the data stored in the memory 8.

[0076] In an estimation step 108, the processing module 6 estimates, from the signals vi' • ■ ■ ' xn, a reference model Uk representative of the reference state of the elongated structure R (therefore the state in which the elongated structure R was during the acquisition of the signals xi' ■ • • ' -Lv).

[0077] As will be seen below, the reference model can be used to evaluate a level of degradation of the state of the elongated structure R compared to the reference state.

[0078] We will now detail an embodiment of this estimation step 108.

[0079] The processing module 6 forms from the N stored signals %i' • • • ' Lv a matrix C of dimensions N*n. The processing module 6 decomposes the matrix C into singular values, in the following manner:

[0080] C = USVT

[0081] where JJ e ^^,3 e r'Vx" and ye R^'.

[0082] S is a diagonal matrix comprising the singular values ​​of the matrix C in a decreasing order, U is a matrix comprising left singular vectors respectively associated with the singular values, and V is another matrix comprising right singular vectors respectively associated with the singular values.

[0083] The reference model Uk is formed by a subset of k vectors selected from the singular vectors on the left of the matrix U or the singular vectors on the right of the matrix V. We therefore have k < n. Preferably, the selected vectors are singular vectors on the left of the matrix U.

[0084] Preferably the value of k is chosen such that a large part of the variance of the signals A i' • • • ' xn is contained in these k selected singular vectors.

[0085] In this case, the matrix C can be approximated by Ck as follows:

[0086] Ck = U kSkVTk

[0087] where er^'.

[0088] The matrices Uk, Sk and Vk constitute portions of the matrices U, S and V discussed previously, limited to the singular values ​​and associated singular vectors with indices ranging from 1 to k.

[0089] Here, the matrix Uk constitutes a reference model formed using left singular vectors. It is understood that the matrix Vk constitutes an alternative model capable of replacing Uk.

[0090] The singular vectors that form the Uk model have the particularity of forming a basis in the vector space considered. Thus, each signal x> having been acquired can be approximated in this basis by:

[0091] xU^

[0092] The parameters p are calculated by minimizing the quadratic error between the estimation of the signals in the basis of the singular vectors and the initially acquired signals, in the following manner:

[0093] p. _ arsmin || xrUkPi || 2 Pi

[0094] The solution to this optimization problem is given by:

[0095] p / jjTyy^T

[0096] The singular vectors model the reference state of the elongated structure R. In other words, if an acquired signal xi comes from an elongated structure R in the reference state, it must be correctly represented in the basis constituted by these vectors. On the other hand, a signal xi acquired later while the structure is in a state different from the reference state will be characterized by an estimation error that can be quantified as follows — x—U kp L

[0097] In other words, the error e< is representative of a difference between the reference state and the state of the elongated structure R during the acquisition of a signal xî.

[0098] As will be seen in more detail below, it is possible to calculate a degradation index of the elongated structure R IDj associated with a signal xi acquired by the acquisition system 2. The degradation index ID[ indicates a level of degradation of the state of the elongated structure R during the acquisition of the signal x' compared to the reference state of the elongated structure R.

[0099] In particular, the degradation index ID{ can be calculated from the error and, in the following manner:

[0100] ID^éfej

[0101] However, it is not necessary to calculate the degradation indices ID^... IDN respectively associated with the signals xi' ■ • • ' xn. We will see later that such indices will, on the other hand, be calculated for other signals acquired subsequently during the monitoring phase.

[0102] In a step 110, the processing module 6 calculates a threshold on the basis of the signals -xi' • • • • xn, The threshold W is a degradation index threshold, in the sense that this threshold is intended to be compared to degradation thresholds ID, which will be calculated later.

[0103] The threshold is for example calculated by the following formula:

[0104] V = gX^h

[0105] where: * h is a chi-square distribution with h degrees of freedom and a level of meaning has, 8 = 9, / 6,

[0106] h = 07 / ¾ • n _ VN Xi, with = 1, 2 • N is the number of first signals acquired, • is a singular value of the matrix S of index j associated with a vector singular on the left not part of the subset of vectors Uk (and possibly VÂ).

[0107] For example, the parameter a is set to the value 0.05.

[0108] During step 110, the processing module can also calculate another threshold r. This threshold r is a variance threshold, and it is used to reveal a break in the elongated structure R during the tracking phase.

[0109] Preferably, the thresholdT is calculated as follows:

[0110] r = ariance^noise)

[0111] where fi is a safety factor Q? S: 1), and where the noise is a signal acquired by the acquisition system 2 in the absence of an ultrasonic wave.

[0112] The learning phase is then complete. 2.2) Follow-up phase

[0113] The follow-up phase is subsequent to the learning phase discussed above.

[0114] The tracking phase comprises successive iterations. By convention, the index Af +1 is assigned to the first iteration implemented during the tracking phase.

[0115] An iteration of index i implemented during the tracking phase comprises the following steps.

[0116] In a step 200, the acquisition system 2 generates a new ultrasonic wave. Step 202 is implemented in the same manner as step 104. The new ultrasonic wave propagates in the elongated structure R according to the methods described previously.

[0117] In a step 202, the acquisition system 2 acquires a new signal representative of this new ultrasonic wave. Step 202 is similar to step 104.

[0118] The new signal -¾ i>N has the same characteristics as the signals ..., xN aCqUis jors phase d'apprentissage. However, this new signal is acquired at a time when the elongated structure R is in a new state which may be identical or different from the reference state in which the elongated structure R was during the learning phase. The new signal xt therefore contains information on this new state of the elongated structure, this information being able to differ from the information contained in the signals ' xn.

[0119] The new signal is stored in memory 8.

[0120] In a step 204, a new temperature Tt i>N associated with the new signal xi. is measured by the measuring system 4. Step 204 is similar to step 106.

[0121] The new temperature T ij>N is stored in the memory 8.

[0122] In a step 206, the processing module 6 compares the new temperature Tt. ,>N and the temperature range [ Tmiw previously stored in memory 8 during the learning phase.

[0123] If the new temperature T, Sc is outside the temperature range [' then a treatment A, which will be detailed later, is implemented, then the process loops back to step 202.

[0124] If on the contrary the new temperature T{ ;>N Sc is in the temperature range [ Tmiw , then the following steps are implemented.

[0125] In a step 208, the processing module 6 determines a degradation index IDi, i>N associated with the new signal signalXi- i>N, this index being determined from the signal xî, and the reference model Uk.

[0126] As indicated previously, the degradation index IDk ï>n indicates a level of degradation of the state of the elongated structure R during the acquisition of the new signal x^ ï>n compared to the reference state.

[0127] The degradation index IDk l>N is determined in step 208 as presented above (after the calculation of the estimate p~^ and the estimation of the error ,>x).

[0128] The higher the value of the degradation index i>N, the more the state of the elongated structure R during the acquisition of the new signal l>N is in principle degraded compared to the reference state.

[0129] Then, in a step 210, the processing module 6 compares the threshold 7 and the degradation index IDk associated with the new signal xî> <>N.

[0130] If i>N then the processing module 6 implements processing B before looping back to step 202.

[0131] If / A, V, then the processing module 6 implements processing C before looping back to step 202.

[0132] With reference to [Fig.3], treatment A comprises the following steps.

[0133] In a step 212, the temperature evolution counter h is incremented by 1 (h = h + 1). Furthermore, the duration of the acquisition period is adjusted to the value P' (short value).

[0134] In a step 214, the processing module 6 calculates the variance V of the new signal é

[0135] In a step 216, the processing module 6 implements a comparison between this variance V and the threshold T determined during the learning phase.

[0136] In a step 218, the processing module generates a result indicating a breakage of the elongated structure R provided that V < T.

[0137] Otherwise (V > r\ the iteration of index * is terminated, and the method loops back to step 202, taking into account the acquisition period (here adjusted to the value P'). In other words, a new acquisition of signal xï+u>n is implemented after a time equal to the acquisition period has elapsed, since the instant of acquisition of signal A'- i>N.

[0138] With reference to [Fig.4], treatment B comprises the following steps.

[0139] In a step 220, the processing module 6 resets the variable m to zero. Furthermore, the duration of the acquisition period is adjusted to the value P.

[0140] In a step 222, the processing module 6 checks whether h > 0. If this is not the case, then the method loops back to the acquisition step 202.

[0141] If the condition h > 0 is verified, then this means that the temperature reached by the elongated structure R during the acquisition of the new acquired signal xv>n has fallen back into the temperature range [ Tmiw Tmax.] after leaving it. More precisely, this reveals that: • h consecutive signals xi- / ^ • • • ' %ii were acquired during h consecutive iterations preceding the iteration with index i; • during these previous iterations, h temperatures of the elongated structure R were measured, all of which are outside the temperature range • from the h consecutive signals Xî-^ ■ • • ' and the reference model U were determined h degradation indices respectively associated with the h consecutive signals which did not cross the threshold 7.

[0142] If these conditions are met (i.e. if h > 0), the processing module 6 sets update the reference model Uk in one go using each of these h consecutive signals • • • ' xi-^ previously acquired.

[0143] It should be noted of course that if h — 1, then the processing module carries out this update on the basis of a single signal, namely the signal xi-h.

[0144] It is understood here that a signal acquired during the tracking phase while the structure is at a temperature outside the range [ Tmiw ] is not immediately used to update the reference model. Rather, we wait for the temperature of the structure to fall back into this range, and we then take advantage of not updating the reference model using a single signal, but if possible several consecutive signals, which is much more economical in terms of computational load. It could be envisaged to carry out h updates of the reference model using a signal in certain embodiments of the production method, but it is less expensive to carry out an update in one go with the h consecutive signals.

[0145] The update 224 of the reference model Uk comprises the same processing as that carried out in the estimation step 108, except that the matrix C is initially enriched with each of the h new consecutive signals acquired while the temperature of the elongated structure R was outside the range [ Tmin, • In other words, the h new consecutive signals are added to the matrix C, and the estimation carried out in step 108 is implemented from the matrix C enriched in this way. This allows the reference model to be refined.

[0146] Furthermore, in a step 226, the processing module 6 updates the temperature range [ Tmjw , so as to extend the temperature range to the h temperatures consecutively measured during the acquisitions of the signals xi-^ • • • ' In other words, it is ensured that after the update, the range includes these h measured temperatures. During this update 226, one and / or the other of the terminals mbv ^max is updated with one of these h measured temperatures.

[0147] In a step 228, the processing module 6 resets the variable h to zero.

[0148] The iteration of index i then ends, and the method loops back to the acquisition step 202.

[0149] With reference to [Fig.5], processing C comprises the following steps.

[0150] In a step 230, the variable m is incremented by 1 (m = m + 1).

[0151] Then, in a step 232, the variable m is compared by the processing module 6 to the predefined value M.

[0152] If m > M then the processing module 6 implements step 216 discussed previously in relation to FIG. 3. If V < t, step 218 is implemented. If on the contrary V > T, the processing module 6 generates during a step 234 a result indicating the presence of a defect in the elongated structure R.

[0153] If m < M then the processing module 6 adjusts at step 236 the duration of the acquisition period to the value P”, then loops back to step 202.

[0154] The value M makes it possible to define a persistence requirement conditioning the generation of the result indicating a defect in the elongated structure R.

[0155] If M is chosen equal to zero, then it is sufficient to have noted that the temperature T, measured at the current iteration of index i is in the temperature range [ Tmiw and that IDj > to trigger the generation 234 of the result indicating the presence of a fault.

[0156] If we choose M > 0, then these conditions must not only be verified at the current iteration i, but also have been verified during M previous iterations to trigger the generation of the result indicating a fault. More precisely, the conditions must be met: • the temperatures Tt_M, ..., T, measured during the respective acquisitions of the M + 1 consecutive signals • • • ' xi were all in the temperature range [Tmirp Tnuix]; • the degradation indices ..., ID, were all above the threshold

[0157] Choosing M > 0 is advantageous, because it avoids certain false alarms caused by aberrant point measurements. We assume that a defect is not transient, but persistent in the elongated structure R.

[0158] In particular, choosing M — 3 offers a good compromise between protection against aberrant one-off measurements and speed of diagnosis. 3) Advantages

[0159] The method described above has many advantages.

[0160] In certain prior methods, false alarms may be caused by the effects of temperature variation. The method proposed in the present disclosure integrates during the monitoring phase temperature variations not observed during the learning phase, which precisely makes it possible to reduce such false alarms.

[0161] In the case where a new temperature variation coincides with the detection of an excessive level of degradation (i.e. when, during an iteration of index i, we have £ [ T ; and IDi > tjY it is not certain that a defect is actually present in the elongated structure R. Indeed, it is possible that a temperature that is too high or too low alters the operation of the acquisition system 2, so that the signal xi is corrupted, resulting in an undue exceeding of the threshold 7 by the associated index 7 / 9.

[0162] The method described above overcomes this problem by relying on the principle of conditional learning. It has in fact been described that the reference model Uk is updated only in the case where the temperature change does not coincide with the detection of an excessive degradation level. To do this, once the system detects a change in temperature, thanks to a temperature sensor of the elongated structure R, it waits for the temperature reached by the elongated structure R to return to its initial variation range before making a decision. Then, as soon as the point at which the temperature returns to the initial variation range is identified, the system calculates the defect index for this point. If it is lower than the threshold, this means that it is only a temperature variation. Therefore, this new temperature variation is integrated into the reference model calculation. Otherwise, it can be assumed that the temperature variation was accompanied by the appearance of a defect. In this case, the persistence test is used to confirm the presence of the defect in the rail.This test consists of not triggering the alarm unless a sufficient number of signals exceed the fault detection threshold.

[0163] Another optional but advantageous feature of the method described above lies in the adjustment of the acquisition period to different values ​​P, P' and P”, with p <P < p. Le principe de cet ajustement est d’accélérer les acquisitions de signaux et les mesures de températures associées lorsque le système 1 est susceptible d’être induit en erreur par une température anormale atteinte par la structure allongée R (ce que l’on détecte en constatant que Ti & £ Tmjw ). Il est avantageux dans cette situation d’accumuler rapidement des nouveaux signaux pour permettre au module de traitement d’arriver rapidement à une prise de décision. En particulier, il est désirable d’obtenir au plus tôt le résultat du test de persistance évoqué plus haut, en ajustant la période à la valeur P” (qui peut être strictement inférieure à P’, ou bien égale à P’).The other principle of this adjustment is to space out signal measurements when we fall back into temperatures already measured, which is a situation less conducive to the occurrence of false alarms. We can then afford to reduce the overall computational load of the process. 2.4) Comparative results

[0164] Figure 6 represents the evolution of the degradation index (starred points *) as a function of time during the implementation of a non-destructive testing method for an elongated rail-type structure R, which is hereinafter called the “control method”. This control method comprises the steps described above with the exception of the steps of updating the model Uk and updating the temperature range [ Tmin> ] • [Fig.6] also represents the evolution of the temperature reached by this same structure over time.

[0165] Figure 6 shows on the left two horizontal segments which delimit tempo- so much the learning period. The tracking phase begins when these two segments become dotted (on the right in Figure 6). The ordinates of these two segments correspond to the limits of the temperature range [ Tmln, ] determined during the learning phase. During the tracking period, the limits of this range do not change given the absence of update steps.

[0166] Figure 6 also shows a horizontal bar whose ordinate corresponds to the value of the degradation threshold V.

[0167] The degradation index curve in Figure 6 shows a peak reaching a maximum on November 17. At first glance, one might assume that the increase in the value of the degradation index reveals a fairly sudden degradation of the state of the elongated structure R. Moreover, this peak crosses the threshold so that the control process generates a result indicating a defect on November 17. However, it is noted that the temperature reached by the elongated structure R drops during the monitoring period to the point of leaving the range [ Tmin > Tmax |, and that this drop in temperature coincides temporally with the peak of the degradation index curve. This coincidence reveals that in reality the increase in the degradation index is not attributable to a real degradation of the state of the elongated structure R, but results from a disturbance in the system implementing the control process, this disturbance being caused by the drop in temperature.Thus, we understand that the result generated by the control process on November 17 constitutes a false alarm. Subsequently, the degradation index increases again very sharply (on November 22) and remains at a plateau. This reveals the appearance of a defect in the elongated structure R.

[0168] Let us now examine Figure 7, which shows the same data as Figure 6, but this time obtained during the implementation of the non-destructive testing method described above (including the steps of updating the reference model Uk and the temperature range [ Tmitr )• We see that the updates carried out when the temperature reached by the elongated structure R leaves the range initially determined during the learning phase have the consequence of preventing the appearance of a peak in the degradation index around November 17. This consequence is logical since the degradation indices are determined according to the reference model C<, and this model was updated using new signals acquired under temperatures that had not yet been observed. On the other hand, the updates carried out in no way prevent the appearance of a sudden increase in the degradation index on November 22.In this regard, it is noted that this increase occurs while the temperature of the structure is within the range of temperatures observed so far. As the degradation index crosses . at this stage the threshold a result reporting a defect in the elongated structure R is correctly generated.

[0169] Ultimately, the comparative study of the results presented in Figures 6 and 7 shows that the updating steps carried out during the process described in this document, which take into account the temperature reached by the elongated structure R during the monitoring period, have the very advantageous effect of eliminating the appearance of false alarms. 3) Implementation variants

[0170] The system and method described above take into account temperatures reached by the elongated structure R, which are measured by the measuring system 4. The temperature is a physical quantity to which the acquisition system 1 is sensitive (in the sense that variations in this physical quantity are likely to disrupt the operation of the acquisition system 2, as indicated previously).

[0171] The system and method described above are, however, generalizable to any physical quantity to which the elongated structure is sensitive. This means that the measurement system 4 can be generalized to a system for measuring any physical quantity to which the acquisition system is sensitive, and that the steps of the method involving temperature values ​​can be generalized to steps involving values ​​of the physical quantity measured by the system 4.

[0172] In certain embodiments, the physical quantity measured is a humidity level (measured in the environment close to the elongated structure) or a pressure experienced by the elongated structure R.

[0173] In other embodiments, the physical quantity may depend on at least two quantities from among: a temperature, a humidity level, a pressure undergone by the elongated structure, or even other quantities, the physical quantity measured is then a combination, for example linear, of quantities measured by sensors of different types forming part of the measuring system 4.

Claims

1. Claims A method of non-destructive testing of an elongated structure such as a rail, the method comprising: • during a first phase, acquisition (104) of reference signals (Ai' • • • ' Av) representative of ultrasonic waves circulating in the elongated structure, the acquisition being implemented by an acquisition system sensitive to a physical quantity, • determination (106) of a range of values ​​([ T ml!V ) of the physical size reached during the first phase, • from the reference signals (xi' • • • ' estimation (108) of a reference model (Uk) representative of a reference state of the elongated structure in which the elongated structure was found during the first phase, • from the reference signals (A i' • • • ' ^ v), calculation (110) of a threshold (^), • during a second phase subsequent to the first phase, acquisition (202) of a new signal (A4 '>aq representative of an ultrasonic wave circulating in the elongated structure, the new signal being acquired while the elongated structure is in a second state, • measurement (204) of a new value (T of the physical quantity during the acquisition of the new signal (¾ • from the new signal (A '• ;>jV) and the reference model ( ^), determination (208) of a degradation index of the elongated structure (® / ) associated with the new signal the degradation index of the elongated structure (^0 associated with the new signal (Xi- indicating a level of degradation of the second state of the elongated structure compared to the reference state of the elongated structure, • update (224) of the reference model (Uk) using the new signal (xï / >N) if at least the following conditions are met: • the new value of the physical quantity (T^ is outside the range of values ​​( [ Tmin, ), and • the degradation index of the elongated structure associated with the new signal Cr'' has not crossed the threshold (^), • update (226) of the range of values ​​so that the range of values ​​includes the new value (T^ of the physical quantity.

2. Method according to the preceding claim, in which: • the acquisition step (202) during the second phase is repeated so that a sequence of signals starting with the new signal (x to ^v) is acquired during the second phase, the sequence comprising h consecutive signals, • the steps of measuring (204) and determining a degradation index (208) are repeated for each of the h consecutive signals, • the updating of the reference model (Uk) is implemented in a single operation using the h consecutive signals if at least the following conditions are met: • during the respective acquisitions of the h consecutive signals of the sequence, h values ​​of the physical quantity were measured which are all outside the range of values ​​( [ Ttniw T^] ), and • from the h consecutive signals of the sequence and the reference model pk),h degradation indices respectively associated with the h consecutive signals which did not cross the threshold (^) were determined.,

3. Method according to any one of the preceding claims, in which the update (224) of the reference model (¼) is triggered after the measurement of a value of the physical quantity which enters for the first time into the range of values ​​([ Tmin, ) since the acquisition of the new signal.

4. A method according to any preceding claim, comprising generating a result indicating the presence of a defect in the elongated structure provided that the degradation index of the elongated structure associated with the new signal has crossed the threshold (V).

5. Method according to the preceding claim, in which: • the acquisition step during the second phase is repeated so that a sequence of signals starting with the new signal is acquired during the second phase, the sequence of signals comprising m consecutive signals, • the generation of the result indicating the presence of a defect in the elongated structure is implemented if the following conditions are met: from the m consecutive signals of the sequence and the reference model (U), h degradation indices respectively associated with the h consecutive signals which have all crossed the threshold (^) have been determined.

6. A method according to any preceding claim, wherein: • the acquisition step during the second phase is repeated during the second phase, the acquisition system using an acquisition period separating the acquisition times of two consecutive signals, • the measurement step is repeated for each of the signals in the sequence, • the acquisition period used by the acquisition system is adjusted to: • a first duration (P) when at least the following conditions are met: • the new value of the physical quantity ( T is in the range of values ​​( • the degradation index of the elongated structure ( / associated with the new signal (on} has not crossed the threshold (^), • a second duration (P') strictly less than the first duration (P), when the new value of the physical quantity (T^ is outside the range of values ​​([7^, • a third duration (P”) less than or equal to the second duration (P'), when at least the conditions following are met: • the new value of the physical quantity (Tj) is in the range of values ​​([TmilP Tmax]), and • the degradation index of the elongated structure (IDi) associated with the new signal (x'> has crossed the threshold (7).

7. Method according to any one of the preceding claims, further comprising the following steps: • calculating a variance (V) of the new signal (x'- ^), • generating a result indicating a breakage of the elongated structure provided that the variance (V) exceeds a predefined variance threshold (r).

8. Method according to any one of the preceding claims, in which the estimation of the reference model (Uk) comprises a singular value decomposition of a matrix C formed from the reference signals, the decomposition being carried out as follows: c= USVT WHERE: • S is a matrix comprising the singular values ​​of the matrix C, • U is a matrix comprising left-hand singular vectors respectively associated with the singular values, • V is a matrix comprising right-hand singular vectors respectively associated with the singular values, the reference model (¼) being formed by a subset of vectors selected from the left-hand singular vectors or from the right-hand singular vectors.

9. Method according to the preceding claim, in which the threshold (7) is calculated by the following formula = gx^ h where: * h is a chi-square distribution with h degrees of freedom and a significance level a, • on ag = 02 / 6Ï eth = 6^16with and ^=1,2 • N is the number of acquired reference signals, • is a singular value of the matrix S of index j associated with singular vectors not selected to form the reference model (¼).

10. A method according to any preceding claim, wherein the physical quantity is or depends on a temperature reached by the elongated structure.

11. Method according to one of the preceding claims, in which the physical quantity is or depends on a humidity level.

12. A method according to any preceding claim, wherein the physical quantity is or depends on a pressure experienced by the elongated structure.

13. System (1) for non-destructive testing of an elongated structure such as a rail, the system (1) comprising: • a signal acquisition system (2) sensitive to a physical quantity, the acquisition system (2) being configured to: • during a first phase, acquire reference signals (xi' • • • ' XN) representative of ultrasonic waves circulating in the elongated structure, • during a second phase subsequent to the first phase, acquire a new signal (i>N) representative of an ultrasonic wave circulating in the elongated structure, the new signal being acquired while the elongated structure is in a second state, • a measurement system (4) configured to: • determine a range of values ​​([ Tmin < Tnwv]) of the physical quantity reached during the first phase, • measure a new value (T0 of the physical quantity during the acquisition of the new signal ( xi,i>N^ • a processing module (6) configured for:, from the reference signals (xi' • • - ' aaz), estimate a reference model (¼) representative of a reference state of the elongated structure in which the elongated structure was found during the first phase, from the first signals (A t' • • • ' ^), calculate a threshold 0), from the new signal (A / - and the reference model (^), determine a degradation index of the elongated structure (^A) associated with the new signal (aa the degradation index of the elongated structure (^A) associated with the new signal indicating a level of degradation of the second state of the elongated structure compared to the reference state of the elongated structure, update the reference model (Uk) using the new signal (¾ if at least the following conditions are met: • the new value (T / ) of the physical quantity is outside the range of values ​​( ïTmbv î'mr])' and • the degradation index of the elongated structure (-^,) associated with the new signal ( did not cross the threshold ('O update the range of values ​​so that the range of values ​​includes the new value (T^ of the physical quantity.

Images