method, device and program for ultrasonic detection of defects in a material

Abstract

The present invention relates to a method for ultrasonic detection of defects in a material (MS), where signals x(n, i, j) are emitted by M emitters of index i, signals x(n, i, j) are received at sampling times n.Te by M receivers of index j, a sampling matrix (AΔ) is formed, having N columns Yn formed by the x(n, i, j) for which a distance between the receiver of index j and the emitter of index i is equal to the gap Δ and lines Xi,j formed by the x(n, i, j), the pair i, j being different from one line Xi,j to the other, a covariance matrix (CΔ) is calculated, projections of the lines Xi,j of the matrix (AΔ) on the K eigenvectors (Vk) corresponding to the K largest eigenvalues ​​(λk), the K projections are subtracted from each line Xi,j, to obtain residual measurement signals x*(n, i, j) defect detection. Figure for abstract: Figure 5

Description

Title of the invention: method, device and program for ultrasonic detection of defects in a material

[0001] The invention relates to a method and a device for ultrasonic detection of defects in a material.

[0002] The field of the invention concerns non-destructive testing of defects by ultrasound, which can be used on molded products of the primary circuits of nuclear power plants, but can also be applied to other components of the nuclear fleet and to other industrial sectors such as aeronautics or naval.

[0003] The present invention is advantageously applicable to complex materials. By complex material is meant any type of material generating parasitic signals which can mask the detection of defects.

[0004] In order to detect potential defects in the material, a common method consists of applying against a surface of the material a multi-element ultrasonic probe, comprising transducers, which is put in turn into transmitter mode and receiver mode of the ultrasonic measurement signals propagating in the material.

[0005] The parasitic signal may, for example, correspond to lateral waves or near-surface waves which can significantly hinder the detection of defects in the first few millimeters of depth of the material. This area of ​​the first few millimeters of depth below the surface of the material is sometimes considered a dead zone because any defects present there may be masked by ambient noise. This parasitic signal may also correspond, for example, to structural noise, this noise resulting from a heterogeneous microstructure of the material. When the wavelength of the ultrasonic waves emitted by the probe is close to the average diameter of the grains of the material, this noise is particularly troublesome for the analysis of the acquisitions, in particular when this noise is higher than the level of the signal reflected by a potential defect.

[0006] The invention seeks to reduce or eliminate the influence of noise in the acquisition of measurement signals.

[0007] A known method for processing this type of ultrasonic measurement signals is the method called Total Focusing Method (TFM). This method produces for each position of the probe an image corresponding to a cross-section of the material under the probe. The presence of parasitic noise (near-surface waves, structural noise) significantly deteriorates the quality of these images, thus compromising the correct detection of potential defects in the material.

[0008] Without possible post-processing on the TFM, the noise generated by surface waves and / or by the heterogeneity of the microstructure can be so significant that it makes the detection of defects difficult or even impossible in certain cases, especially at shallow depths.

[0009] The parasitic noise observed on the images from the TFM presents a spatial inhomogeneity: it varies significantly depending on the location of the points. In particular, it tends to be higher near the emitter / receiver array (surface wave effect).

[0010] Document FR-A-3085481 discloses a method for detecting and characterizing defects in a heterogeneous material by ultrasound, providing for post-processing the image obtained, for example by the TFM method, using the following statistics: a measurement of the central tendency of the amplitude focused at the point probed on different probe positions, a measurement of the function representing the variability of the amplitude focused at the point probed on different probe positions. Another focusing method taught by document FR-A-3085481 is the plane wave imaging method, designated PWI (from the English "Plane Wave Imaging"), where the different configurations are distinguished from each other by different delays applied to the emission of ultrasonic waves by the emitting transducers by exciting all the transducers uni-sequentially.

[0011] However, these statistics used in the method known from document FR-A-3085481 assume that there are healthy areas (free from defects) in the image, so that these healthy areas can be learned beforehand to constitute representative areas or reference areas of the inspected material. These measurements will then make it possible to normalize the noise and to identify any amplitude deviations synonymous with defects. The statistics produced on the healthy areas correspond to a learning phase of the material being studied. This learning phase can only be carried out if healthy and representative areas are available, serving as reference areas. In the absence of such healthy and representative areas, serving as reference areas, it would be necessary to implement a preliminary step to the exploitation of the method of document FR-A-3085481, and therefore organize additional tests, which increases the overall duration of the measurements.

[0012] An aim of the present invention is to provide a method and a device for ultrasonic detection of defects in a material, which overcome the drawbacks mentioned above and which save on such a reference zone when it is not available.

[0013] To this end, a first object of the invention is a method for ultrasonic detection of defects in a material, characterized in that the method comprises the steps following: a) ultrasound waves are successively emitted against a surface of the material by M emitting ultrasonic transducers of index i of a multi-element probe, where i is a first natural integer ranging from 1 to M and where M is a second prescribed natural integer greater than or equal to 2, we receive, at sampling times n.Te by M ultrasonic receiver transducers of index j of the multi-element probe, measurement signals x(n, i, j) which are representative of the amplitude of the ultrasound propagated in the material, where n is a third natural integer ranging from 1 to N, where N is a fourth prescribed natural integer greater than or equal to 2, where Te is a prescribed sampling period and where j is a fifth natural integer ranging from 1 to M, b) a sampling matrix is ​​formed by a calculator, for at least one prescribed deviation A, which is positive or zero, having N columns Yn, the N columns Yn, for n ranging from 1 to N, being formed by the set of measurement signals x(n, i, j) and corresponding to the N sampling instants n.Te, each column Yn having for the sampling instant n.Te the set of measurement signals x(n, i, j) for which a distance between the receiving ultrasonic transducer of index j and the emitting ultrasonic transducer of index i is equal to the prescribed gap A, which is identical for the N columns Yn, the sampling matrix having lines Xij formed by the set of measurement signals x(n, i, j), for which the index i is identical in each line Xjj and the index j is identical in each line Xjj, the pair i, j being different from one line X ij to another, c) a covariance matrix is ​​calculated by the calculator from the sampling matrix, the covariance matrix being a square and symmetric matrix of dimension N x N, d) the calculator calculates p eigenvectors and p eigenvalues ​​associated with the eigenvectors for the covariance matrix, where p is a prescribed sixth natural integer, greater than or equal to 2 and is a prescribed maximum number of calculated eigenvectors and calculated eigenvalues, less than or equal to N, e) we calculate by the projection calculator ^Pro.i'k of the lines Xjj of the matrix ij sampling on the K eigenvectors corresponding to the K largest eigenvalues, where K is a selected number less than the maximum number p of calculated eigenvectors and calculated eigenvalues, f) the calculator subtracts from each of the lines Xjj of the sampling matrix the projections of this line Xjj onto the K eigenvectors, to obtain residual fault detection lines X*^ formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*^ to another.

[0014] Thanks to the invention, the parasitic signals in which the possible defects of the inspected material were embedded are attenuated or eliminated, which makes it possible to better detect these defects. Thus, the invention does not require the learning phase taught by document FR-A-3085481.

[0015] Embodiments of the invention are described below, which can be applied to the ultrasonic defect detection method according to the invention, to the ultrasonic defect detection device according to the invention and to the ultrasonic defect detection computer program according to the invention.

[0016] According to one embodiment of the invention, e) the projections ^PiC'jA are calculated by the calculator, for k ranging from 1 to K, of the ij rows Xjj of the sampling matrix on the K eigenvectors corresponding to the K largest eigenvalues, where K is a selected number less than the maximum number p of calculated eigenvectors and calculated eigenvalues, where k is a seventh natural integer ranging from 1 to K, f) the calculator subtracts from each of the lines Xjj of the sampling matrix the K projections ^Proi>k , for k ranging from 1 to K, of this line X^, to obtain residual fault detection lines X*^ formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*^ to another.

[0017] According to one embodiment of the invention, the method further comprises the following step: g) the calculator performs post-processing of material defect detection from the residual defect detection measurement signals x*(n, i, j).

[0018] According to one embodiment of the invention, the defect detection post-processing comprises an algorithm for focusing on the residual defect detection measurement signals x*(n, i, j) to generate an image.

[0019] According to one embodiment of the invention, the focusing algorithm is an al- all-point focusing algorithm, the all-point focusing algorithm comprising a step of calculation by the calculator of an indicator I*(w) for probed positions w in a cross-section of the material as follows: * y A / yi A / $ I (W) = L i=lL j = iX" (r( w, z, j ), z, j ) where t(w, i, j) corresponds to a time of path for a signal, which was emitted by the transmitting ultrasonic transducer of index i, which was reflected at the probed position w and which was received by the receiving ultrasonic transducer of index j, where t(w, i, j) corresponds to one of the sampling instants n.Te and where n is calculated between 1 and N, and a step of formation, by the computer, of the image, for which the probed positions w correspond to positions of pixels of the image, the value of the pixels of the image at the positions w being equal to the indicator I*(w).

[0020] According to one embodiment of the invention, the focusing algorithm is a full-point focusing algorithm, the full-point focusing algorithm comprising a step of calculation by the computer of an indicator I*(w) for probed positions w in a cross-section of the material in the following manner: I"02" f(w) = ) where corresponds to a travel time for a signal, which has been emitted by the transmitting ultrasonic transducer of index i, which has been reflected at the probed position w and which has been received by the receiving ultrasonic transducer of index j, where t(w, i, j) corresponds to one of the sampling instants n.Te and where n is calculated between 1 and N, and a step of formation, by the computer, of the image, for which the probed positions w correspond to positions of pixels of the image, the value of the pixels of the image at the positions w being equal to the indicator I*(w) and where g is a prescribed function.

[0022] According to one embodiment of the invention, the method further comprises the following step:

[0023] h) the computer performs bilateral filtering of the image.

[0024] According to an embodiment of the invention, the computer (CAL) calculates the integer K, for which 2K > m + 2s and + j tn + 2s, where ( 2* ) < æ sn denotes the eigenvalues ​​for a seventh natural integer k ranging from 1 to N, m is the average of the N eigenvalues ​​(2^.) < k < N, s is the standard deviation of the N eigenvalues ​​( 2^ ) <

[0025] According to one embodiment of the invention, the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed regularly relative to the surface of the material.

[0026] According to one embodiment of the invention, the M ultrasonic transducers transmitters of index i are respectively part of M ultrasonic transmitting-receiving units located in respectively M distinct prescribed positions in the multi-element probe, and the M ultrasonic receiving transducers of index j are respectively part of the M ultrasonic transmitting-receiving units.

[0027] According to one embodiment of the invention, the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed in a plane.

[0028] According to one embodiment of the invention, the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed along at least one rectilinear axis.

[0029] According to an embodiment of the invention, the computer calculates the residual measurement signals x*(n, i, j) for detecting defects for several prescribed deviations A, which are different from each other.

[0030] According to an embodiment of the invention, the computer calculates the residual measurement signals x*(n, i, j) for detecting defects for the prescribed deviations A corresponding to all the combinations of the M emitting ultrasonic transducers of indices i with the M receiving ultrasonic transducers of indices j.

[0031] According to one embodiment of the invention, the M emitting ultrasonic transducers of index i of the multi-element probe are merged with the M receiving ultrasonic transducers of index j of the multi-element probe.

[0032] According to another embodiment of the invention, the M emitting ultrasonic transducers of index i of the multi-element probe are distinct from the M receiving ultrasonic transducers of index j of the multi-element probe.

[0033] A second object of the invention is a device for ultrasonic detection of defects in a material, characterized in that the device comprises:

[0034] a multi-element probe, comprising M emitting ultrasonic transducers of index i, capable of successively emitting ultrasound against a surface of the material, where i is a first natural integer ranging from 1 to M and where M is a second prescribed natural integer greater than or equal to 2,

[0035] the multi-element probe comprising M receiving ultrasonic transducers of index j, capable of receiving, at sampling times n.Te, measurement signals x(n, i, j) which are representative of the amplitude of the ultrasound propagated in the material, where n is a third natural integer ranging from 1 to N, where N is a fourth prescribed natural integer greater than or equal to 2, where Te is a prescribed sampling period and where j is a fifth natural integer ranging from 1 to M,

[0036] the device comprising a calculator, which is configured to - form for at least one prescribed deviation A, which is positive or zero, a matrix sampling, having N columns Yn,

[0037] the N columns Yn, for n ranging from 1 to N, being formed by the set of measurement signals x(n, i, j) and corresponding to the N sampling instants n.Te, each column Yn having for the sampling instant n.Te the set of measurement signals x(n, i, j) for which a distance between the receiving ultrasonic transducer of index j and the emitting ultrasonic transducer of index i is equal to the prescribed gap A, which is identical for the N columns Yn,

[0038] the sampling matrix having lines Xjj formed by the set of measurement signals x(n, i, j), for which the index i is identical in each line Xjj and the index j is identical in each line Xjj, the pair i, j being different from one line X ij to another, - calculate a covariance matrix from the sampling matrix, the covariance matrix being a square and symmetric matrix of dimension N x N,

[0039] - calculate p eigenvectors and p eigenvalues ​​associated with the eigenvectors for the covariance matrix, where p is a prescribed sixth natural integer, greater than or equal to 2 and is a prescribed maximum number of computed eigenvectors and computed eigenvalues, less than or equal to N,

[0040] - calculate projections ^-Pr°j>k of the lines Xij of the sampling matrix on the K eigenvectors corresponding to the K largest eigenvalues, where K is a selected number less than the maximum number p of calculated eigenvectors and calculated eigenvalues,

[0041] - subtract from each of the lines Xjj of the sampling matrix the projections ^proj.k this line Xjj on the K eigenvectors, to obtain residual lines X*jj fault detection lines formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*jj to another.

[0042] According to an embodiment of the invention, a calculator, which is configured to -calculate the projections ^P10^, for k ranging from 1 to K, of the lines Xjj of the matrix kj sampling on the K eigenvectors corresponding to the K largest eigenvalues, where K is a selected number less than the maximum number p of calculated eigenvectors and calculated eigenvalues, where k is a seventh natural integer ranging from 1 to K,

[0043] - subtract from each of the lines Xjj of the sampling matrix the K projections ^.proj,k , for ranging from 1 to K, of this line Xjj, to obtain residual lines X*jj fault detection lines formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*jj to another.

[0044] A third object of the invention is a computer program for ultrasonic defect detection, comprising code instructions for implementing the ultrasonic defect detection method as described above, when executed by a computer.

[0045] The invention will be better understood on reading the description which follows, given solely by way of non-limiting example with reference to the figures below of the attached drawings.

[0046] [Fig-1] represents a schematic perspective view of a detection device by ultrasound of defects according to an embodiment of the invention.

[0047] [Fig.2] represents a schematic cross-sectional view of a device for ultrasonic detection of defects according to one embodiment of the invention.

[0048] [Fig.3] represents a schematic cross-sectional view of a device for ultrasonic detection of defects according to one embodiment of the invention.

[0049] [Fig.4] represents a schematic cross-sectional view of a device for ultrasonic detection of defects according to one embodiment of the invention.

[0050] [Fig.5] represents a flowchart of an ultrasonic detection method of defects according to an embodiment of the invention.

[0051] [Fig.6] represents measurement signals acquired during the detection process by ultrasound of defects according to an embodiment of the invention.

[0052] [Fig.7] represents an image obtained without implementing the method and the device for detecting defects according to an embodiment of the invention.

[0053] [Fig.8] represents an image obtained by implementing the method and the device for detecting defects according to an embodiment of the invention from the measurement signals of [Fig.7].

[0054] [Fig.9] represents an image obtained by implementing the method and the device for detecting defects according to an embodiment of the invention which further comprises the application of bilateral filtering.

[0055] [Fig. 10] is a graph representing the signal-to-noise ratio as a function of a chosen component number parameter of the method and the fault detection device according to an embodiment of the invention.

[0056] [Fig. 11] represents an image obtained from the same configuration as in [Fig.7] by the method and the device for detecting defects according to an embodiment of the invention with use of the selection criterion of the number of components and addition of bilateral filtering.

[0057] [Fig. 12] represents an image obtained from the configuration used for the graph of [Fig. 10] without having implemented the method and the device for detecting defects according to an embodiment of the invention.

[0058] [Fig. 13] represents an image obtained from the configuration used for the graph of [Fig. 10] by the method and the fault detection device according to an embodiment of the invention with use of the selection criterion of the number of components and addition of bilateral filtering.

[0059] The ultrasonic defect detection method, the ultrasonic defect detection device 100 and the computer program implementing this method are described below with reference to FIGS. 1 to 5. The defect detection device 100 comprises a multi-element probe 10, comprising ultrasonic transducers 14, 15 which can be ultrasound emitters and / or receivers. The steps of the ultrasonic defect detection method are described with reference to [Fig. 5]. During a first step E1, the multi-element probe 10 is placed in a certain position z on the surface S of the material MS to be inspected. For example, the multi-element probe 10 may comprise a coupling medium located between the surface S of the material MS to be inspected and the ultrasonic transducers 14, 15, to allow propagation of ultrasound between the surface S of the material MS to be inspected and the ultrasonic transducers 14, 15.In one embodiment of the invention, this coupling medium may be an integral part of the multi-element probe 10 and be integral with the ultrasonic transducers 14, 15 and may comprise, for example, a gel contained in a container integral with the ultrasonic transducers 14, 15, the multi-element probe 10 being against or in contact with the surface S of the material MS to be inspected in this case.In another embodiment of the invention, this coupling medium is not an integral part of the multi-element probe 10 and is not integral with the ultrasonic transducers 14, 15 and can be added between the surface S of the material MS to be inspected and the ultrasonic transducers 14, 15, this coupling medium being able to be for example a height of water present between the surface S of the material MS to be inspected and the ultrasonic transducers 14, 15, for example in the case where the material MS and the multi-element probe 10 are immersed in water, the multi-element probe 10 being at a short distance from the surface S of the material MS to be inspected in this case.

[0060] The MS material to be inspected can be any type of material, in particular coarse-grained materials, such as for example coarse-grained steels.

[0061] The multi-element probe 10 comprises M emitting ultrasonic transducers 14, these emitting ultrasonic transducers 14 respectively having an index i (which is a first natural integer) ranging from 1 to M, where M is a second prescribed natural integer greater than or equal to 2. The multi-element probe 10 comprises M ul- transducers 15 receivers having respectively an index j (which is a fifth natural integer) ranging from 1 to M.

[0062] During a second step E2, subsequent to the first step E1, the M emitting ultrasonic transducers 14 of index i of the multi-element probe 10 successively emit an ultrasonic signal SI against the surface S of the material MS at respective successive emission times IE;. In response to each ultrasonic signal SI emitted by each emitting ultrasonic transducer 14 of index i at the respective emission instant IE;, (and before the respective emission instant IEi+i of the following emitting ultrasonic transducer 14 of index i+1), the M receiving ultrasonic transducers 15 of index j receive during the second step E2 the ultrasonic measurement signals x(n, i, j) which are representative of the amplitude of the ultrasound S2 propagated in the material MS, for j ranging from 1 to M. Each receiving ultrasonic transducer of index j receives these ultrasonic measurement signals x(n, i, j) at instants n.Te sampling time (after the respective emission time IE; of the emitting ultrasonic transducer 14 of index i and before the respective emission time IEi+i of the next emitting ultrasonic transducer 14 of index i+1), where n is a third natural number from 1 to N, where N is a prescribed fourth natural number greater than or equal to 2 and where Te is a prescribed sampling period (inverse of a prescribed sampling frequency). The M emitting ultrasonic transducers 14 of index i of the multi-element probe 10 and the M receiving ultrasonic transducers 15 of index j can have the first transmission-reception configuration, which will be described below. The sampling times n.Te are defined to within a constant relative to an initial time.

[0063] According to one embodiment of the invention, each ultrasonic transducer 14, 15 can alternately play the role of transmitter or receiver. Each ultrasonic transducer 14, 15 can be put into an ultrasound transmitter mode or into an ultrasound receiver mode. In this case, the M transmitting ultrasonic transducers 14 of index i of the multi-element probe 10 are merged with the M receiving ultrasonic transducers 15 of index j of the multi-element probe 10.

[0064] According to another embodiment of the invention, the M transmitting ultrasonic transducers 14 of index i of the multi-element probe 10 are distinct from the M receiving ultrasonic transducers 15 of index j of the multi-element probe 10.

[0065] According to an embodiment of the invention, the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are regularly distributed in the multi-element probe 10, as shown by way of non-limiting example in FIGS. 2 and 3. For example, the M emitting ultrasonic transducers 14 of index i are part of M ultrasound transmitting-receiving units 13 located in M ​​separate prescribed positions in the multi-element probe 1, and the M receiving ultrasonic transducers 15 of index j are respectively part of these M ultrasound transmitter-receiver units 13, as shown by way of non-limiting example in Figures 2 and 3. Of course, any other distribution of the ultrasonic transducers 14, 15 can be provided.

[0066] According to an embodiment of the invention, the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are distributed in a plane P, for example parallel to the surface S of the material MS and possibly on the surface S of the material MS or at a non-zero distance from the surface S of the material MS, as shown by way of non-limiting example in FIGS. 2 to 4.

[0067] According to an embodiment of the invention, the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are distributed along one (or more) rectilinear axes 16, as shown by way of example in FIGS. 2 and 3. In this case, the probe 10 can be a multi-element bar where the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are aligned along this rectilinear axis 16.

[0068] According to one embodiment of the invention, the plane P in which the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are located is parallel to the plane of the surface S of the material MS examined.

[0069] According to an embodiment of the invention, one or more or all of the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j may be at a distance from the surface S of the material MS.

[0070] According to another embodiment of the invention, the plane P in which the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are located is inclined relative to the surface S, and one or more or all of the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j may be at a distance from the surface S of the material MS.

[0071] According to another embodiment of the invention, the plane P in which the M emitting ultrasonic transducers 14 of index i and the M receiving ultrasonic transducers 15 of index j are located is at a non-zero distance from the surface S, and may be parallel or inclined relative to the surface S of the material MS.

[0072] According to an embodiment of the invention, the M emitting ultrasonic transducers 14 of index i are distributed according to M prescribed coordinates iL spaced from each other by the same prescribed pitch L not zero relative to each other (this coordinate may be an abscissa iL along one (or more) rectilinear axes 16 with the rectilinear pitch L between them, or may be an angle iL around another axis in the case of the M emitting ultrasonic transducers 14 of index i distributed around this other axis with the angular pitch L between them, or others), and / or the M receiving ultrasonic transducers 15 of index j are distributed according to M prescribed coordinates jL spaced apart from each other by a prescribed non-zero pitch L relative to each other (this coordinate may be an abscissa jL along one (or more) rectilinear axes 16 with the rectilinear pitch L between them, or may be an angle iL around another axis in the case of the M receiving ultrasonic transducers 15 of index j distributed around this other axis with the angular pitch L between them, or others), as shown by way of non-limiting example in Figures 2 and 3. Of course, the M transmitting ultrasonic transducers 14 of index i may not be distributed according to the same prescribed non-zero pitch L relative to each other and / or the M receiving ultrasonic transducers 15 of index j may not be distributed according to the same prescribed non-zero pitch L relative to each other.

[0073] For a given position of the probe 10 relative to the surface S of the material MS, a 3-dimensional matrix of measurement signals x(n, i, j) is available, where n corresponds to a number of discrete time steps, i is the index of the emitting ultrasonic transducer 14 and j is the index of the receiving ultrasonic transducer 15.

[0074] During a third step E3, subsequent to the second step E2, a CAL computer forming part of the fault detection device 100 forms a sampling matrix Aa, having N columns Yn. The sampling matrix Aa groups together all the measurement signals x(n, i, j) acquired for n ranging from 1 to N having the same distance difference A between the receiving ultrasonic transducers 15 of index j and the transmitting ultrasonic transducers 14 of index i. The N columns Yn are formed by the measurement signals x(n, i, j) and correspond to the N sampling instants n.Te of the M receiving ultrasonic transducers 15. The CAL computer can be part of the probe 10 or is linked or connected to the probe 10.It may be provided as a CAL calculator, for example, one or more calculators and / or one or more computers), and / or one or more processors and / or one or more servers and / or one or more machines, which may be programmed in advance by a pre-recorded computer program for implementing the method and may include one or more permanent memories, on which this program is pre-recorded. The CAL calculator automatically executes the steps of the fault detection method. Another calculator connected to the probe 10 or integrated into the probe 10 can record the acquisitions, which this other calculator or the CAL calculator can then process by the method according to the invention.

[0075] Each column Yn is formed from the measurement signals x(n, i, j) for which a distance d(i, j) between the receiving ultrasonic transducer 15 of index j and the transmitting ultrasonic transducer 14 of index i is equal to a prescribed gap A, which is identical for the N columns Yn (and therefore equal to the distance d(i', j') between the receiving ultrasonic transducer 15 of index j' and the transmitting ultrasonic transducer 14 of index i' for the pair (i,j) different from the pair (i',j')). The computer is prescribed CAL the gap A for each matrix Aa. This gap A can be positive or zero. We therefore have:

[0076] d(i, j) = d(i', j') = A for the pair (i,j) different from the pair (i',j') in the matrix Aa.

[0077] The lines Xjj of the matrix Aa are formed from the measurement signals x(n, i, j) for n ranging from 1 to N, for which the index i is identical in each line Xjj and the index j is identical in each line X^j, the pair i, j being different from one line Xjj to another.

[0078] The CAL calculator therefore forms the sampling matrix Aa, which is defined by the following equations:

[0079] Aa — (Yf .. ■ Y„... Y,v)

[0080]

[0081]

[0082]

[0083]

[0084] Y^ y„ = YïV = A a = Aa - x(l, i. x(l, i', x(n, i, x(n, i', ' x(N, ii' / x(l, i) U(U i' k J j)' ni jH ni , j) , j') , j) , j1) ■ ■ ■ x(n, i, j) ■ ■ ■ x(N, i, j) \ -x(n, f, j')-- x(N, i', j') /

[0085] j) ■ ■ ■ x(n, i, j) ■ ■ ■ x(N, i, j) )

[0086]

[0087] The gap A is constant in each sampling matrix Aa. In the absence of defects in the material MS, the signals x(n, i, j) with the same gap A have the same spatial and temporal characteristics. This is illustrated in [Fig.6], where several measurement signals x(n, i, j) for n belonging to an interval included in [1, ..., N] with different pairs i,j between the receiving ultrasonic transducer 15 of index j and the transmitting ultrasonic transducer 14 of index i but with the same distance gap A = 10 between these receiving ultrasonic transducers 15 of index j and these transmitting ultrasonic transducers 14 of index i have been represented as a non-limiting example and where these measurement signals x(n, i, j) are substantially identical for different positions of these receiving ultrasonic transducers 15 of index j and these emitting ultrasonic transducers 14 of index i. The abscissa axis corresponds to the time n in number of samples, and the ordinate axis corresponds to the amplitude x(n, i, j) of the received signal. The frequency of the ultrasonic waves SI emitted by the emitting transducers 14 of index i is, by way of non-limiting example, 5 MHz in [Fig.6].

[0088] For example, in each column Yn of the sampling matrix Aa associated with A = 5L, we will find, for n ranging from 1 to N:

[0089] - the signal x(n, 1, 6) emitted by the transmitting ultrasonic transducer 14 of index i=1 and received by the receiving ultrasonic transducer 15 of index j=6, - the signal x(n, 2, 7) emitted by the transmitting ultrasonic transducer 14 of index i=2 and received by the receiving ultrasonic transducer 15 of index j=7, - the signal x(n, 3, 8) emitted by the transmitting ultrasonic transducer 14 of index i=3 and received by the receiving ultrasonic transducer 15 of index j=8, - etc...

[0090] During steps E4 to E6 described below and subsequent to the third step E3, the CAL calculator implements a principal component analysis algorithm (PCA for short) on the sampling matrix Aa.

[0091] During the fourth E4 after the third step E3, the CAL calculator calculates the covariance matrix CA corresponding to the sampling matrix Aa described above. The coefficient crq> of the r-th row and the q-th column of the covariance matrix CA is equal to: cr>q = cov(Yr, Yq) for r being a natural integer ranging from 1 to N and q being a natural integer ranging from 1 to N, where Yr, Yq designate the columns of the sampling matrix Aa. The covariance matrix CA is calculated as a function of the sampling matrix Aa according to the following equation, for n ranging from 1 to N:

[0092] Aa = (Yj ...Y„... Yv)

[0093] / cov^, Y] ) = var^) ■■■ cov(Y], Y„) cov(Y15 Y^) CA= : : : cov(Yv Yj) ■ ■ ■ cov(Ya„ Yn)■■■ cov( YA, YA,) = var(YjV);

[0094] where var denotes the variance of a column and cov denotes the covariance between two columns. The covariance matrix CA is square, of dimension N x N and symmetric.

[0095] During the fifth step E5 after the fourth step E4, the calculator CAL calculates p eigenvectors Vk and p eigenvalues ​​Xk associated with the eigenvectors Vk for the covariance matrix CA corresponding to the sampling matrix Aa, for k ranging from 1 to p. The number p is a prescribed sixth natural integer, greater than or equal to 2 and is a prescribed maximum number of calculated eigenvectors Vk (and is a prescribed maximum number of calculated eigenvalues ​​Xk). By example, p=N. In another example, we could have p <N. Dans un autre exemple, on pourrait avoir p<N (on peut calculer moins de vecteurs propres que de nombre N de colonnes de la matrice CA de covariance). Pour p=N, les vecteurs propres Vk forment une base : toute ligne de la matrice Aa peut être décomposée comme combinaison linéaire des N vecteurs propres Vk pour k allant de 1 à N. In the sixth E6 after the fifth step E5, the CAL calculator calculates projections ^prok k for k going from 1 to K of the rows Xjj of the matrix Aa Æ fi sampling on the K eigenvectors Vk corresponding to the K largest eigenvalues ​​Xk of the covariance matrix CA corresponding to the sampling matrix Aa, with K <p. Les valeurs propres de la décomposition en vecteurs propres d’une matrice de covariance sont réelles positives, du fait que cette matrice de covariance est carrée et symétrique. On prescrit au calculateur CAL ou on détermine par le calculateur CAL le nombre K, qui est un nombre inférieur au nombre maximum p de vecteurs propres Vk et de valeurs propres Xk. Pour ce faire, le calculateur peut ordonner les valeurs propres Xk de la matrice CA de covariance correspondant à la matrice Aa d’échantillonnage et sélectionner les K valeurs propres Xkles plus grandes.

[0096] The measurement signals x(n, i, j) can contain several pieces of information: information associated with surface waves, structural noise, but also possibly information associated with a defect in the MS material, a defect that we wish to detect. Complex MS materials generate noise because of the different wave scattering phenomena. What is more, noise is omnipresent in the measurement signals x(n, i, j), and conversely, a defect is not always present. Even if there is a defect, then it does not represent a large amount of information on all the signals targeted. Thus, the information contained in the K eigenvectors with large eigenvalues ​​Xk is that of the predominant parasitic noise (typically the surface wave), while the information contained in the other eigenvectors with small eigenvalues ​​corresponds to the possible defects / artifacts present in the signals.A property of the decomposition into eigenvectors Vk and eigenvalues ​​Xk is that the K eigenvectors Vk corresponding to the K largest eigenvalues ​​Xk will correspond to predominant information, while the eigenvectors Vk associated with the small eigenvalues ​​Xk will correspond to information more submerged in this predominant information. The number K is therefore the number of components to be retained from the decomposition into eigenvectors of the covariance matrix CA corresponding to the sampling matrix Aa.

[0097] During the seventh E7 after the sixth step E6, the calculator CAL subtracts from each of the lines Xjj of the sampling matrix Aa the projections yproP k for k going from 1 to K of this line Xjj on the K eigenvectors Vk cor-i, j corresponding to the K largest eigenvalues ​​Xk of the covariance matrix CA corresponding to this sampling matrix Aa. The result of this operation is, following the same formalism as the sampling matrix Aa, residual defect detection lines X*ij formed by the residual defect detection measurement signals x*(n, i, j) (called signals x* below) for n ranging from 1 to N, for which the index i is identical in each residual defect detection line X*ij and the index j is identical in each residual defect detection line X*ij, the pair i, j being different from one residual defect detection line X*ij to another. This removes a certain amount of noise from the signals, so that the defect information of the material MS takes over. Thus, we subtract from each line vector Xjj the projection of itself onto the K eigenvectors Vk corresponding to the K i,} largest eigenvalues ​​Xk. The number K is therefore the number of components removed. The CAL calculator therefore calculates the residual fault detection lines X*^ corresponding to the residual fault detection measurement signals x*(n, i, j) for n ranging from 1 to N, which are defined by the following equations:

[0098]

[0099]

[0100] \. = (x(l, i, j) ■■■ x(n, i, j) ■■■ x(N, i, j)) M 1 ' j)-" x % L j)'"X*(N, i, j)) ^proj, k CS( |c projected vector of Xjj on the k-th eigenvector Vk:

[0101] vnk l, J \ l, A *}

[0102] where the operator ()T denotes transposition,

[0103] yP^ is the sum of the projected vectors of the Xjj on the K eigenvectors Vj, ..., Vk , ..., VK for k ranging from 1 to K:

[0104] _ y K ^P'o.vk », jk= II, j

[0105] where Vk is a row vector. Therefore, at the seventh E7, the computer CAL subtracts from each of the rows X^ of the sampling matrix Aa the sum ypr°j of the projections yprG|' * for k going from 1 to K of this line X^ on the K eigenvectors Vk corresponding to the K largest eigenvalues ​​Xk of the covariance matrix CA corresponding to this sampling matrix Aa, to obtain the residual line X*^ of fault detection.

[0106]

[0107] By convention, vectors are represented with a capital letter.

[0108] The residual measurement signals x*(n, i, j) for detecting defects are therefore freed from a certain amount of noise to reveal the small variations representing the defects in the MS material.

[0109] According to an embodiment of the invention, the CAL calculator calculates the residual measurement signals x*(n, i, j) for detecting faults for several prescribed deviations A, which are different from each other.

[0110] According to an embodiment of the invention, the CAL calculator calculates the residual measurement signals x*(n, i, j) for detecting defects for all the prescribed deviations A corresponding to all the combinations of the M emitting ultrasonic transducers 14 of indices i with the M receiving ultrasonic transducers 15 of indices j.

[0111] According to one embodiment of the invention, K can be equal from one prescribed gap A to the other and therefore from one sampling matrix Aa to the other.

[0112] According to another embodiment of the invention, K can be different from one prescribed gap A to another and therefore from one sampling matrix Aa to another.

[0113] According to an embodiment of the invention, the computer CAL carries out, during an eighth E8 subsequent to the seventh step E7, a post-processing of detection of defects of the material MS from the residual measurement signals x*(n, i, j) of detection of defects.

[0114] According to an embodiment of the invention, the computer CAL performs, during the eighth step E8, as post-processing for defect detection, a focusing algorithm on the residual measurement signals x*(n, i, j) for defect detection. Of course, the post-processing for detecting defects of the material MS may be other than the focusing algorithm, or may even be absent, the eighth step E8 being optional.

[0115] Embodiments of the invention of this focusing algorithm, of the all-point focusing type (FTP for short, or TFM in English), are described below during step E8.

[0116] According to an embodiment of the invention, the focusing algorithm is carried out during step E8 by the CAL computer on the residual defect detection measurement signals x*(n, i, j). The CAL computer calculates an image in which each pixel of the image represents a probed point w of the MS material with which a focused amplitude is associated for said probed point w.

[0117] By way of non-limiting example, during an ultrasonic shot at a position z of the multi-element probe 10, one (or more) emitting transducer 14 emits ultrasonic waves SI which penetrate into the material MS at its surface S, then propagate in the material MS, before being received by a receiving transducer 15. In order to illustrate the propagation of the ultrasonic waves SI and S2 in the material, on the [Fig.4] a first path T1 constituting a short path for the ultrasonic waves SI and S2, which are diffracted by the defect DEF in the direction of the receiving transducer 15, and a second path T2 constituting a long path for the ultrasonic waves SI and S2, which are reflected by another surface S' of the material M, distant from its surface S, in the direction of the defect DEF and then join the receiving transducer 15.

[0118] In one approach, the different configurations are distinguished from each other by transmitter or receiver functions fulfilled by different transducers 14, 15.

[0119] For example, in a first transmitting-receiving configuration, a first transducer 14 (or a first set of transducers 14) is individually excited with a pulsed electrical signal in order to emit SI ultrasound. These ultrasounds propagate in the material, and are then received by all the transducers 14, 15 (or by a second set of transducers 15). Then, at the same probe position z, another transducer 14 (or another first set of transducers 14) is individually excited with a pulsed electrical signal in order to emit SI ultrasound. These SI ultrasounds propagate in the material, and are then received by all the transducers 14, 15 (or by another second set of transducers 15). Preferably, each of the transducers 14, 15 emits ultrasound in at least one probe configuration at a position z.Typically, each of the transducers 14, 15 is in turn the only transmitting transducer, while all the transducers 14, 15 receive the ultrasounds. There are then as many ultrasonic shots as there are transducers 14, 15 in the first transmission-reception configuration. Of course, this first transmission-reception configuration is not limiting and other transmission-reception configurations of the transducers 14 and 15 can be provided.

[0120] According to an embodiment of the invention, the all-point focusing algorithm comprises, during step E8, a step of calculation by the calculator CAL of an indicator I*(w) for each probed position w in the following manner: [01211

[0122] where t(w, i, j) corresponds to the travel time (expressed in number of samples) for a signal emitted by the transmitting ultrasonic transducer of index i, reflected at the probed position w and received by the receiving ultrasonic transducer of index j, t(w, i, j) corresponds to one of the sampling instants n.Te and where n is calculated between 1 and N, and a step of forming, by the computer CAL, an image I*, for which w represents the probed positions in a cross-section of the material MS and corresponds to pixel positions of the image I*, the value of the pixels of the image I* at positions w being equal to the indicator I*(w). Thus, w represents a probed position in a cross-section of the material and corresponds to a position of a pixel of the I* image, the value of a pixel of the I* image at position w being equal to the indicator I*(w). The final I* image is produced when the I* indicators for the positions w of all the pixels of the I* image have been calculated.

[0123] Thus, t(w, i, j) can be the travel time for an ultrasonic signal:

[0124] emitted by transducer i, reflected by a point of the supposed DEF defect located at position w, and captured by the transducer].

[0125] But t(w, i, j) can also be the travel time for a signal:

[0126] emitted by transducer i, reflected by the background S' of the material MS, reflected by a point of the supposed DEF defect located at position w, and captured by transducer j.

[0127] These travel times t(w, i, j) are calculated by the CAL calculator from the wave speeds which depend on the type of propagation (transverse waves, longitudinal waves). Mode conversions can also be considered during the different reflections.

[0128] The indicator I*(w) being calculated for a cross-section of the MS material as a function of the position z of the probe 10, the CAL calculator will finally obtain an image I* which will subsequently be called imaging or TFM image.

[0129] The TFM formula can be generalized during step E8 in the form: 101301 A w) = £"£.,£( X-( t(W. i, j).i. j ) )where »COTresP°"d“ travel time (expressed in number of samples) for a signal emitted by the transmitting ultrasonic transducer of index i, reflected at the probed position w and received by the receiving ultrasonic transducer of index j, where t(w, i, j) corresponds to one of the sampling instants n.Te and where n is calculated between 1 and N, and a step of forming, by the computer CAL, an image I*, for which w represents the probed positions in a cross-section of the material MS and corresponds to pixel positions of the image I*, the value of the pixels of the image I* at positions w being equal to the indicator I*(w), where g is a prescribed function. Thus, w represents a probed position in a cross-section of the material and corresponds to a position of a pixel of the image I*, the value of a pixel of the image I* at position w being equal to the indicator I*(w). The final image I* is produced when the I* indicators for the positions w of all pixels in the I* image have been calculated..

[0131] Several variants of the TFM during step E8 are possible depending on the choice of g. The most common choices of g are as follows:

[0132] g(x) = x, g(x) is different from x,

[0138] RSB = 20*log - g depends on the probed position w and / or the propagation speed of the wave, - g can also be the absolute value of the signal or the modulus of the signal analytical.

[0133] The method and device 100 for detecting defects according to the invention allow a gain in decibels, which facilitates the detection of defects, in particular near the surface S of the material MS.

[0134] Indeed, [Fig.7] represents a TFM image obtained solely from the signals x(n, i, j) without implementing the method and the device 100 for detecting defects according to the invention. This TFM image of [Fig.7] contains a defect close to the surface S, which is however difficult to identify because of the phenomena explained above.

[0135] By applying the method and the device 100 for detecting defects according to the invention, we obtain, from the signals x*(n, i, j) described above, the TFM image of [Fig.8], with initially the parameter K fixed at 4.

[0136] A defect about 2.5 mm deep is highlighted in [Fig.8], which was drowned in the very present noise close to the surface S of [Fig.7].

[0137] In one embodiment of the invention, an additional step E9 of post-processing the TFM image can be applied by the CAL computer after the eighth step E8. This additional step E9 of post-processing the TFM image can comprise bilateral filtering applied to the image. [Fig.9] shows the image obtained by applying bilateral filtering to the image of [Fig.8]. This filtering makes it possible to improve the visual quality of the image as well as its SNR (Signal to Noise Ratio). The SNR is a quantitative indicator of the quality of such an image. The SNR is expressed in decibels (dB) and is calculated for example in the following manner: jmax s defect I MX I noise /

[0139] Where Inffut corresponds to the maximum intensity of the pixels in the defect area,

[0140] corresponds to the maximum intensity of the pixels in the noise area, i.e. the entire area except the defect area. A negative SNR indicates that the noise is of greater intensity than the defect, while a positive SNR indicates that the noise is of less intensity than the defect. This filter is effective only because the image quality in Figure 8 is sufficient (i.e., there is a good distinction between the defect area and the rest).

[0141] The example illustrated in Figures 7 and 8 achieved an SNR gain of 18.1 dB (by increasing the SNR from 4.7 in [Fig.7] to 22.8 in [Fig.8]). The image in [Fig.9] has an SNR of 33.2.

[0142] A summary of the performances on different transducer configurations and for different fault depths are given below.

[0143] [Tables 1] Configuration 1 (defect depth = 5 mm) Configuration 2 (defect depth = 5 mm) Configuration 3 (defect depth = 10 mm) Configuration 4 (defect depth = 10 mm) Configuration 5 (defect depth = 15 mm) SNR from x signals (without x*) -1.2 -25.7 4.3 - 37.2 -4.4 SNR from x* signals and K=4 23.3 41.6 24.0 34.1 28.1 SNR from x* signals and K=4 and with bilateral filtering 29.1 56.7 35.3 50.3 42.5

[0144] The graph in [Fig. 10] shows the SNR as a function of K for the same configuration. Curve C1 shows the SNR obtained from the signals x (without x*). Curve C2 shows the SNR obtained from the signals x*, without applying bilateral filtering of the final image. Curve C3 shows the SNR obtained from the signals x*, with applying bilateral filtering of the final image. The SNR of the raw signal, represented by curve C1 formed by a horizontal line, is -1.0. This means that initially, the maximum intensity of the noise is greater than the maximum intensity of the defect. By applying the method and the device 100 for detecting defects according to the invention, with more or less components removed (depending on K), the SNR is significantly improved. The gain in SNR can be increased with the addition of bilateral filtering in additional post-processing E9 of the TFM image.

[0145] In other embodiments of the invention, the additional step E9 of post-processing the TFM image is not present.

[0146] The larger K is, the more components we remove, and by removing information, we will end up removing that which is associated with possible DEF defects. It is for this reason that curves C2 and C3 are increasing up to a certain threshold, then decreasing.

[0147] The optimal value of K is not the same for all configurations, because it depends on the nature of the MS material and therefore on the acquisitions.

[0148] According to one embodiment of the invention, the number K is prescribed to the CAL calculator.

[0149] According to an embodiment of the invention, K can be determined by the CAL calculator using one of the methods for automatic selection of the number of components known from the literature from the results obtained during steps E3 to E5. Thus, it is guaranteed that the choice of K is linked to and dependent on the particular case studied. In other words, the nature of the MS material to be inspected influences the choice of K.

[0150] According to an embodiment of the invention, the CAL calculator could implement a method for calculating K from the eigenvalues ​​(2a), and / or the eigenvectors (Vk), and / or the covariance matrix CA, and / or the matrix AA, and / or the acquisitions x(n,i,j).

[0151] According to an embodiment of the invention, the CAL calculator retains from among the set of N eigenvalues ​​(2a ) N resulting from the decomposition into eigenvectors of the covariance matrix CA, only those which are greater than m + 2s, with m the average of the N eigenvalues, and s the standard deviation of the N eigenvalues. Consequently, the eigenvalues ​​( 2^ ) < k < N being arranged in decreasing order: 2 । > 22 ... 4: ÀN, the integer K to be retained by the calculator CAL is the one for which 2K > m + 25 and 2^ + j < m + 2s.

[0152] According to an embodiment of the invention, in the case where p=N, the mean m and the standard deviation s are calculated by the CAL calculator in the following way, since we have access to all the eigenvalues: 101531 101541 .=LJ ) - w J

[0155] These calculations assume that the number p of calculated eigenvalues ​​is equal to the maximum number N of eigenvalues.

[0156] According to another embodiment of the invention, the CAL calculator calculates the mean m and the standard deviation s of the set of eigenvalues ​​directly from the covariance matrix CA. We first recall a property linking the eigenvalues ​​(2k ) < and the covariance matrix CA: 101571

[0158] where:

[0159] trace correSpOnd to the operator, for a square matrix, denoting the sum of the diagonal terms of the matrix,

[0160] l is an integer greater than or equal to 1 corresponding to the power to which the covariance matrix CA is raised in the left-hand member, and to the power to which the eigenvalues ​​(scalars) are raised in the right-hand member.

[0161] Using this property with l — 1, we obtain for the mean m the following result:

[0162] _ _1 yn trace^C^ m~ N k=iÂk~ 7y

[0163] Using this property with / = 2, we obtain the following result for the standard deviation s:

[0164] s = = Nm2 ) =

[0165] The CAL calculator can therefore calculate the mean m and the standard deviation s directly from the covariance CA matrix, and this before having calculated the eigenvalues ​​and the eigenvectors of the covariance CA matrix. The CAL calculator can then calculate only the p=K eigenvalues ​​and eigenvectors.

[0166] [Fig. 11] shows the image of the same configuration as in [Fig.7] using the selection criterion of the number K of components and adding bilateral filtering. The SNR is 35.7.

[0167] [Fig. 12] shows the TFM image of the configuration used for the graph of [Fig. 10] obtained only from the signals x(n, i, j) without implementing the method and the device 100 for detecting faults according to the invention. (RSB=-1.0).

[0168] Figure 13 shows the TFM image obtained for the same example as before using the transformed acquisitions x*(n,i,j), with selection of the number K of components for each A and addition of bilateral filtering. The method used for the choice of K in this example is the one with m+2s described above. The SNR is 46.6, which is slightly larger than the best case of the graph shown in Figure 10. This improvement comes from the fact that for the graph in Figure 10, the K is the same for all A .

[0169] According to one embodiment of the invention, the CAL calculator recalculates the eigenvectors and the number K of components to be removed for each new control of an MS material.

[0170] According to another embodiment of the invention, the CAL calculator records the eigenvectors and the number K of components to be removed from a given MS material, in order to be able to reuse them the following times on this given MS material. The calculation time will automatically be greatly reduced the following times.

[0171] According to another embodiment of the invention, it may be interesting not to consider all of the acquired samples x(n, i, j) for n ranging from 1 to N, but to restrict oneself to a given time zone, for example:

[0172] x( [h0, n0+l, ...,22^, / ,7)

[0173] with n0 (respectively nj the first (respectively the last) sample considered for each signal x(n, i, j). This makes it possible to concentrate on a time zone of interest, corresponding to the depth of the defect sought, and to reduce the calculation time.

[0174] According to one embodiment of the invention, the defect detection method is executed for several positions z (different from each other) of the multi-element probe 10 on the surface S of the material MS to be inspected.

[0175] Of course, the embodiments, features, possibilities, variants and examples described above can be combined with each other or selected independently of each other.

Claims

Claims

1. Method for ultrasonic detection of defects in a material (MS), characterized in that the method comprises the following steps: a) ultrasounds are successively emitted against a surface (S) of the material (MS) by M emitting ultrasonic transducers of index i of a multi-element probe (10), where i is a first natural integer ranging from 1 to M and where M is a second prescribed natural integer greater than or equal to 3, at sampling times n.Te are received by M receiving ultrasonic transducers of index j of the multi-element probe (10), measuring signals x(n, i, j) which are representative of the amplitude of the ultrasounds propagated in the material (MS), where n is a third natural integer ranging from 1 to N, where N is a fourth prescribed natural integer greater than or equal to 2, where Te is a prescribed sampling period and where j is a fifth natural integer ranging from 1 to M, b) a computer (CAL) forms, for at least a prescribed gap A,which is positive or zero, a sampling matrix (Aa), having N columns Yn, the N columns Yn, for n ranging from 1 to N, being formed by the set of measurement signals x(n, i, j) and corresponding to the N sampling instants n.Te, each column Yn having for the sampling instant n.Te the set of measurement signals x(n, i, j) for which a distance between the receiving ultrasonic transducer of index j and the emitting ultrasonic transducer of index i is equal to the prescribed gap A, which is identical for the N columns Yn, the sampling matrix (Aa) having rows Xij formed by the set of measurement signals x(n, i, j), for which the index i is identical in each row Xij and the index j is identical in each row Xjj, the pair i, j being different from one row Xjj to another, c) a matrix (CA) is calculated by the calculator (CAL) of covariance from the sampling matrix (Aa),the covariance matrix (CA) being a square and symmetric matrix of dimension N x N, d) the calculator (CAL) calculates p eigenvectors (Vk) and p eigenvalues ​​(Xk) associated with the eigenvectors (Vk) for the covariance matrix (CA), where p is a prescribed sixth natural integer, greater than or equal to 2 and is a prescribed maximum number of eigenvectors, calculated (Vk) and calculated eigenvalues ​​(Xk), less than or equal to N, e) we calculate by the calculator (CAL) projections xPrûJ'k ^es ''8ncs «.j Xjj of the sampling matrix (Aa) on the K eigenvectors (Vk) corresponding to the K largest eigenvalues ​​(Xk), where K is a selected number less than the maximum number p of calculated eigenvectors (Vk) and calculated eigenvalues ​​(Xk), f) the calculator (CAL) subtracts from each of the lines Xjj of the sampling matrix (Aa) the projections ^Pr°jA of this line Xij on the K eigenvectors (Vk), to obtain residual fault detection lines X*^ formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*^ to another.

2. Method according to claim 1, characterized in that it further comprises the following step: g) the computer (CAL) carries out a post-processing of detection of defects in the material (MS) from the residual measurement signals x*(n, i, j) of detection of defects.

3. Method according to claim 2, characterized in that the defect detection post-processing comprises an algorithm for focusing on the residual defect detection measurement signals x*(n, i, j) to generate an image.

4. Method according to claim 3, characterized in that the focusing algorithm is a full-point focusing algorithm, the full-point focusing algorithm comprising a step of calculation by the computer (CAL) of an indicator I*(w) for probed positions w in a cross-section of the material (MS) in the following manner: corresponds to a travel time for a signal, which was emitted by the emitting ultrasonic transducer of index i, which was reflected at the probed position w and which was received by the receiving ultrasonic transducer of index j, where t(w, i, j) corresponds to one of the sampling instants n.Te and where n is calculated between 1 and N, and a step of formation, by the computer (CAL), of the image (I*), for which the probed positions w correspond to pixel positions of the image (I*), the value of the pixels of the image (I*) at positions w being equal to the indicator I*(w).

5. Method according to claim 3, characterized in that the focusing algorithm is a full-point focusing algorithm, the full-point focusing algorithm comprising a step of calculation by the computer (CAL) of an indicator I*(w) for probed positions w in a cross-section of the material (MS) in the following manner: / \w) = E 'w=1 EZ(f(w, / ,7), / ,7) ) corresponds to a travel time for a signal, which was emitted by the transmitting ultrasonic transducer of index i, which was reflected at the probed position w and which was received by the receiving ultrasonic transducer of index j, where t(w, i, j) corresponds to one of the instants n.Sampling step and where n is calculated between 1 and N, and a step of formation, by the calculator (CAL), of the image (I*), for which the probed positions w correspond to pixel positions of the image (I*), the value of the pixels of the image (I*) at the positions w being equal to the indicator I*(w) and where g is a prescribed function.

6. Method according to any one of claims 3 to 5, characterized in that it further comprises the following step: h) bilateral filtering of the image is carried out by the computer (CAL).

7. Method according to any one of the preceding claims, characterized in that the calculator (CAL) calculates the integer K, for which > m + 2s and +1 - m + 2s, where ( ^ ) i < k < n denotes the eigenvalues ​​for a seventh natural integer k ranging from 1 to N, m is the average of the N eigenvalues ​​( ), < k < N, s is the standard deviation of the N eigenvalues ​​(2^) t < jV-

8. Method according to any one of the preceding claims, characterized in that the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed regularly relative to the surface (S) of the material (MS).

9. Method according to any one of the preceding claims, characterized in that the M transmitting ultrasonic transducers of index i are part of M ultrasonic transmitting-receiving units located in M ​​separate prescribed positions in the multi-element probe, and the M receiving ultrasonic transducers of index j are respectively part of the M ultrasonic transmitting-receiving units.

10. Method according to any one of the preceding claims, characterized in that the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed in a plane (P).

11. Method according to any one of the preceding claims, characterized in that the M emitting ultrasonic transducers of index i and the M receiving ultrasonic transducers of index j are distributed along at least one rectilinear axis (16).

12. Method according to any one of the preceding claims, characterized in that the computer (CAL) calculates the residual measurement signals x*(n, i, j) for fault detection for several prescribed deviations A, which are different from each other.

13. Method according to claim 12, characterized in that the computer (CAL) calculates the residual measurement signals x*(n, i, j) for detecting defects for the prescribed deviations A corresponding to all the combinations of the M emitting ultrasonic transducers of indices i with the M receiving ultrasonic transducers of indices j.

14. Method according to any one of claims 1 to 13, characterized in that the M transmitting ultrasonic transducers of index i of the multi-element probe (10) are merged with the M receiving ultrasonic transducers of index j of the multi-element probe (10).

15. Method according to any one of claims 1 to 13, characterized in that the M transmitting ultrasonic transducers of index i of the multi-element probe (10) are distinct from the M receiving ultrasonic transducers of index j of the multi-element probe (10).

16. Device for ultrasonic detection of defects in a material (MS), characterized in that the device comprises: a multi-element probe (10), comprising M emitting ultrasonic transducers of index i, capable of successively emitting ultrasound against a surface (S) of the material (MS), where i is a first natural integer ranging from 1 to M and where M is a second prescribed natural integer greater than or equal to 3, the multi-element probe (10) comprising M receiving ultrasonic transducers of index j, capable of receiving, at sampling instants n.Te, measurement signals x(n, i, j) which are representative of the amplitude of the ultrasound propagated in the material (MS), where n is a third natural integer ranging from 1 to N, where N is a fourth prescribed natural integer greater than or equal to 2, where Te is a prescribed sampling period and where j is a fifth natural integer ranging from 1 to M, the device comprising a calculator (CAL), which is configured to - form for at least one prescribed deviation A, which is positive or zero, a sampling matrix (Aa), having N columns Yn, the N columns Yn, for n ranging from 1 to N, being formed by the set of measurement signals x(n, i, j) and corresponding to the N sampling instants n.Te, each column Yn having for the moment n.Te of sampling the set of measurement signals x(n, i, j) for which a distance between the receiving ultrasonic transducer of index j and the emitting ultrasonic transducer of index i is equal to the prescribed gap A, which is identical for the N columns Yn, the sampling matrix (Aa) having rows Xij formed by the set of measurement signals x(n, i, j), for which the index i is identical in each row Xij and the index j is identical in each row Xij, the pair i, j being different from one row Xij to another, - calculate a covariance matrix (CA) from the sampling matrix (Aa), the covariance matrix (CA) being a square and symmetric matrix of dimension N x N, - calculate p eigenvectors (Vk) and p eigenvalues ​​(Xk) associated with the eigenvectors (Vk) for the covariance matrix (CA), where p is a prescribed sixth natural integer, greater than or equal to 2 and is a prescribed maximum number of calculated eigenvectors (Vk) and calculated eigenvalues ​​(Xk), less than or equal to N, - calculate projections ^Pro.bk of the rows X^ of the matrix (Aa) ij sampling on the K eigenvectors (Vk) corresponding to the K largest eigenvalues ​​(Xk), where K is a selected number less than the maximum number p of calculated eigenvectors (Vk) and calculated eigenvalues ​​(Xk), - subtract from each of the lines X^ of the sampling matrix (Aa) the projections ^-P'^A of this line X^ on the K ij eigenvectors (Vk), to obtain residual lines X*ij of fault detection formed by a set of residual fault detection measurement signals x*(n, i, j), for which the index i is identical in each residual fault detection line X*^ and the index j is identical in each residual fault detection line X*^, the pair i, j being different from one residual fault detection line X*^ to another.

17. Device according to claim 16, characterized in that the M emitting ultrasonic transducers of index i of the multi-element probe (10) are merged with the M receiving ultrasonic transducers of index j of the multi-element probe (10).

18. Device according to claim 16, characterized in that the M transmitting ultrasonic transducers of index i of the multi-element probe (10) are distinct from the M receiving ultrasonic transducers of index j of the multi-element probe (10).

19. Computer program for ultrasonic defect detection, comprising code instructions for implementing the ultrasonic defect detection method according to any one of claims 1 to 15, when executed by a computer (CAL).

Images