TOMOGRAPHIC ANALYSIS PROCEDURE

Description

Technical field of the invention

[0001] The invention relates to a method for tomographic analysis of a mechanical part. The part is, for example, part of an aircraft, and more specifically of a turbomachine, such as part of a fan casing, a fan blade, or part of a fixed blade structure. State of the prior art

[0002] It is known to inspect structural mechanical parts by tomography methods, for example radiographic or acoustic, in order to detect the presence of possible defects on the surface and inside the part.

[0003] These methods are reliable, non-invasive and allow for the inspection of the interior of parts, which makes it possible to quickly determine the condition of the part and decide whether it should be put back into operation or replaced.

[0004] A tomographic measurement consists of scanning an observed object, here a mechanical part, using a beam of waves, and measuring the beam transmitted in all directions in order to reconstruct a three-dimensional image of the object.

[0005] There figure 1 illustrates a tomography device 1 during the inspection of a mechanical part 20.

[0006] The tomography device 1 comprises at least one emission device 3, configured to emit an incident beam 5 of wave pulses in the direction of the part 20, for example radiofrequency waves, X-rays or even acoustic waves, and at least one receiver 7 capable of capturing a transmitted beam 9 of waves, arranged on either side of the part 20.

[0007] The part 20 is generally mounted on a support 11 rotating around an axis A, in order to be observed from all directions during the tomographic measurement.

[0008] The part 20 observed by tomography may be made of a composite material, as shown in detail in the figure 2 , which includes several distinct phases comprising different materials.

[0009] Such a composite material comprises, for example, weft fibers 21 and warp fibers 22 woven together on a weaving plane P, or several weaving planes P superimposed along a thickness direction Z. Such a weave is designated as two-dimensional. These fibers 21, 22 are also designated by the term strands, and these two terms are used interchangeably hereinafter.

[0010] The fibers of the composite material can, alternatively, follow a three-dimensional weave, with fibers extending in three distinct directions.

[0011] The fibers 21 and 22 may be carbon fibers, glass fibers, synthetic polymer fibers (for example of the “Kevlar” type (registered trademark)), or a mixture of several types of fibers. They are embedded in a matrix 23 comprising for example one or more polymers and / or resins, which solidify to form the final part 20.

[0012] The amplitudes of the waves transmitted from the different observation angles are translated into gray levels in a processing device 13, and a digital analysis makes it possible to reconstruct the volume of the part 20. The processing device 13 comprises a computer. The gray levels obtained depend, among other things, on the density of the materials crossed, and therefore vary between the different phases of the composite material.

[0013] Tomography analysis then provides a visual of the exterior and interior of the part. This type of inspection has the major advantage of allowing visualization through the thickness of the material and the part being studied, while being non-destructive and reliable.

[0014] A human operator can then assess the quality of the part, by identifying the different phases of the part and studying the weaving pattern, in order to detect the possible presence of an anomaly or damage.

[0015] Indeed, many of the mechanical properties of the final piece depend on the good performance of the woven fibers, which can be assessed by several indicators. These include the warp / weft ratio and the fiber volume ratio. These parameters can also vary within the weaving process, with strands of different sizes or otherwise.

[0016] In the following, the terms "non-conformity", "anomaly", "damage", "damage" or equivalent are used interchangeably to designate a part of the part where the mechanical properties of the part are locally degraded compared to those of a part in good condition.

[0017] It may also be the presence of a foreign body in the material, such as a void (porosity) or material that has arrived by mistake in the part during its shaping. Such an anomaly may justify removal and replacement of the part when its mechanical strength is compromised, or, depending on the case, may not prevent the proper functioning of the part. In the case of a composite material, it may also be an area where the fibers 21, 22 are stretched or broken.

[0018] It is important to note that the analysis of the tomography image is still currently left to the human operator, who must separate the fibers and the matrix and study the weave manually. This step generates a significant delay because the three-dimensional image is complex and extensive, which makes its inspection tedious.

[0019] The document Sencu et al: "Generation of micro-scale finite element models from synchrotron X-ray CT images for multidirectional carbon fibre reinforced composites", COMPOSITES PART A (2016), discloses a method for automatic analysis of the weaving of a composite part. Presentation of the invention

[0020] The invention aims to overcome these drawbacks by proposing an analysis method automatically providing information on the distribution of fibers in a radiographic image. Such information allows, for example, a digital reconstruction of the volume of the observed part.

[0021] To this end, the invention relates to a method for tomographic analysis of a composite part comprising a matrix and fibers embedded in the matrix, the method comprising the following steps: acquiring at least one two-dimensional image of the part by means of a tomography device, generating a plurality of characteristic shapes of the fibers and, optionally, for each characteristic shape, a convolution mask comprising a plurality of copies of the characteristic shape, calculating, for each of the characteristic shapes, a convolution product of the two-dimensional image with the characteristic shape or with the corresponding convolution mask, and obtaining a convolution image, detecting a position in the two-dimensional image corresponding to a global maximum on all the convolution images obtained for each of the characteristic shapes, and assigning a fiber center to said position, marking as processed a region of the image arranged around the fiber center and corresponding to the characteristic shape for which the maximum was obtained, and iterating the calculation steps,detection and marking on unprocessed regions of the image.

[0022] Such a method makes it possible to automatically detect fibers in a two-dimensional image resulting from a section of a tomographic image, and thus to collect information on the positions, dimensions and orientations of the fibers in order to observe and / or model the part.

[0023] The method may comprise a step of preprocessing the image, comprising at least one of filtering an average value of the image, and removing edges of the image. Such a filtering step makes it possible to overcome parasitic effects resulting from the tomographic acquisition, such as for example an overall gradient on the image.

[0024] Filtering can be done using a median Gaussian type filter.

[0025] The step of acquiring the at least one two-dimensional image may comprise acquiring a three-dimensional image of the part using the tomography device and making at least one planar section in the three-dimensional image to obtain each two-dimensional image.

[0026] Such a step makes it possible to generate several two-dimensional images oriented according to the different directions of interest of the part, from the tomography results.

[0027] The characteristic shapes of the fibers can be ellipses.

[0028] Such a geometry allows efficient detection and obtaining useful information on the orientation and arrangement of the fibers, while remaining simple enough to limit the computational time.

[0029] Among the generated characteristic shapes, at least two of the characteristic shapes may have different dimensions from each other.

[0030] Such a feature makes it possible to take into account variations in the dimensions of the fibers in the part.

[0031] In the case where the characteristic shapes are ellipses, the dimensions are for example the length of the major axis and the length of the minor axis of each ellipse.

[0032] Among the generated characteristic shapes, at least two of the characteristic shapes may have different inclinations from each other relative to a reference direction of the two-dimensional image.

[0033] Such a feature makes it possible to take into account variations in the orientation of the fibers in the room and thus improve detection.

[0034] Among the generated convolution masks, at least two convolution masks may comprise characteristic shapes arranged with different spacings between said two convolution masks.

[0035] Such a feature makes it possible to take into account local variations in fiber arrangement and thus improve detection.

[0036] The spacings are for example measured between the centers of two neighboring fibers.

[0037] The part may be a turbomachine casing part, or a compressor rotor, stator or fan blade of a turbomachine.

[0038] The invention also relates to a computer program product comprising instructions allowing, when executed on a computer, the implementation of the method. Brief description of the figures

[0039] [ Fig. 1 ] there figure 1 is a schematic side view of a tomography device during the implementation of a method according to the invention, [ Fig. 2 ] there figure 2 is a schematic detail view of a part made of woven composite material, [ Fig. 3 ] there figure 3is a schematic view of a step of preparing images of the method according to the invention, [ Fig. 4 ] there figure 4 is a view of examples of characteristic shapes of strands used in the method according to the invention, [ Fig. 5 ] there Figure 5 is a view of examples of masks comprising a plurality of characteristic shapes of the figure 4 , And [ Fig. 6 ] there figure 6 is a view of the progressive detection of strands in an image by the method according to the invention. Detailed description of the invention

[0040] A tomographic analysis method according to the invention will now be described. This method implements the analysis device 1 described previously in relation to the figure 1 , to observe a room 20.

[0041] The part 20 is for example made of composite material, as shown in the figure 2 .

[0042] The composite material comprises weft fibers 21 extending in a weft direction X and warp fibers 22 extending in a warp direction Y, embedded in a matrix 23. The average distance between two neighboring weft fibers 21 or warp fibers 22 is of the order of 2 mm.

[0043] A thickness of the part 20, measured along a thickness direction Z, is for example between 5 and 25 mm, in particular close to 10 mm for a part 20 comprising four to eight planes of superimposed weaving planes.

[0044] The weft 21 and warp 22 fibers may be of identical or different materials (glass, carbon or “Kevlar” (registered trademark)).

[0045] The matrix 23 comprises at least one organic polymer and / or at least one resin.

[0046] Woven composite materials can be considered as orthotropic materials, that is, materials having three planes of symmetry in their internal microstructure.

[0047] Alternatively, the composite material may comprise weft 21 and warp 22 fibers, as well as fibers extending in the thickness direction Z, thus forming a three-dimensional weave.

[0048] The method aims to obtain information relating to the distribution of the fibers 21, 22 in the matrix 23. For this, the method aims to obtain the positions of the centers of the sections of the fibers 21, 22 and the dimensions and orientations of said sections of the fibers 21, 22 in at least one two-dimensional image corresponding to a section of the part 20 in a predetermined plane.

[0049] The method comprises a first step of acquiring at least one three-dimensional image of the part 20, by means of the tomography device 1.

[0050] An incident beam 5 of wave pulses is emitted by the emission device 3, in the direction of the part 20. The incident beam 5 is for example a beam of radiofrequency waves.

[0051] The waves pass through the part 20, and a transmitted beam 9 from the part 20 is picked up by the receiver 7. The intensity distribution of the transmitted beam 9 obtained is converted into a two-dimensional grayscale image of the part 20 by the processing device 13. The part 20 is rotated by means of the support 11, and two-dimensional images of the part 20 are acquired from all directions.

[0052] A three-dimensional image 30, shown on the figure 3 , of the part 20 in gray levels, or tomographic image, is then reconstituted by image processing using the processing device 13, the image comprising a plurality of voxels each having a respective gray level.

[0053] The acquisition step then comprises the preparation of at least one two-dimensional image by producing, for each two-dimensional image, a section of the three-dimensional image 30 in a respective plane P1, P2, P3, as shown in the figure 3

[0054] Each cutting plane P1, P2, P3 can be, for example, perpendicular to the weft direction, the warp direction or the thickness direction.

[0055] The two-dimensional images obtained are substantially rectangular images, comprising a plurality of pixels each having a respective gray level.

[0056] The method advantageously comprises a step of preprocessing each two-dimensional image making it possible to improve the obtaining of results by the analysis method.

[0057] Preprocessing includes filtering an average gray level value of the image, using a median Gaussian filter.

[0058] Indeed, the image may include a parasitic grayscale gradient resulting from the tomography acquisition, which can disrupt subsequent analysis. Filtering on the average value makes it possible to remove this gradient for the rest of the process.

[0059] On the other hand, applying such a filter to the image can make the edges of the image unusable, which must then be removed. Preprocessing can then include a step of cropping the edges of the image and recentering the resulting image.

[0060] The method then comprises a step of generating a plurality of characteristic shapes 40 of the fibers 21, 22, used to detect the sections of the fibers in each two-dimensional image.

[0061] The characteristic shapes 40, represented on the figure 4, are approximations of the section of a fiber 21, 22 in the image plane. They have, for example, elliptical shapes and each have a center C, a major axis A1 and a minor axis A2.

[0062] The position of an ellipse is defined by its center C, its dimensions by the lengths of the major axis A1 and the minor axis A2, and its orientation by an angle α formed between the major axis A1 and a reference direction of the image, for example the thickness direction Z in the example shown.

[0063] The dimensions of the characteristic shapes 40 can be determined by direct observation on the image of a certain number of fibers 21, 22, for example between 5 and 10, and by identification of average values ​​and associated standard deviations.

[0064] Another method of determining the dimensions of characteristic shapes is to start from the knowledge of the initial fibers 21, 22 before the formation of the composite part 20 and their inclusion in the matrix 23. The size of the composite shapes can then be obtained by applying a scale factor between the actual dimensions of the part and the size of the pixels of the image.

[0065] Based on this information, about ten characteristic shapes 40 are generated with dimensions varying from one to the other.

[0066] A greater or lesser number of characteristic shapes 40 can be determined, depending on the available computing time and the desired precision.

[0067] The orientations of the characteristic shapes 40 can theoretically be arbitrary. However, practice shows that the orientations correspond to angles α forming a distribution mainly between -30° and 30° around a median orientation.

[0068] Thus, the method comprises the identification of this median orientation on the image by observing a certain number of sections of fibers 21, 22, and taking into account orientations distributed around this median value at constant angular pitch, for example 10°, between -30° and 30° around the median orientation.

[0069] The combination of these orientations applied to the previously determined multiple-dimensional characteristic shapes 40 gives, for example, between 50 and 100 different characteristic shapes 40 that can be searched for in the image.

[0070] Depending on the desired duration of the calculations, the angular step can be increased or decreased to obtain more or less characteristic shapes 40.

[0071] For each characteristic shape, the method comprises preparing at least one convolution mask 50, 51, 52, as shown Figure 5 , comprising a plurality of such identical characteristic shapes 40, arranged together to form a pattern.

[0072] The arrangement of the characteristic shapes 40 can be varied from one convolution mask 50, 51, 52 to another, by varying the spacing between the characteristic shapes 40 or their respective positions.

[0073] For example, the first mask 50 comprises shapes oriented with a zero angle α, the second mask 51 inclined shapes with α=30° and the third mask 52 shapes oriented with a zero angle α and a reduced distance between the shapes according to the thickness direction Z.

[0074] The method then comprises a step of calculating, for each of the convolution masks 50, 51, 52, a convolution product of the two-dimensional image with the convolution mask, and obtaining a convolution image resulting from this calculation.

[0075] Alternatively, the calculation of the convolution product can be done with a single characteristic form 40.

[0076] The convolution calculation is then faster, but the results obtained are less sensitive to the variability of the strands.

[0077] The convolution calculation is performed for each of the 40 characteristic shapes, i.e. several dozen times, and as many convolution images are obtained.

[0078] The convolution images obtained by the calculation for each of the characteristic shapes 40 are then compared with each other, so as to detect a global maximum value on all the convolution images.

[0079] The position in the two-dimensional image of the pixel giving said maximum value corresponds to the center of a strand, and the characteristic shape for which the maximum value was obtained provides the dimensions and orientation of the strand section in the cutting plane of the image. The corresponding area therefore corresponds to a detected strand and is marked as processed and excluded from subsequent iterations of the convolution calculation.

[0080] The steps of computing convolution products, detecting the maximum on the obtained convolution images and marking the corresponding area as a new strand are iterated on the unprocessed regions of the two-dimensional image, in order to detect the strands of the image one by one. The progressively obtained results are represented on the figure 6 .

[0081] The most obviously visible strands in the two-dimensional image are detected first, then the process progressively detects the less obvious strands.

[0082] This detection order simplifies the analysis of the most complex areas after excluding previously detected strands.

[0083] The detected strands are advantageously visually displayed on the two-dimensional image, so that an operator can manually stop the detection process when he considers a sufficient number of strands detected.

[0084] Alternatively, a process stopping criterion may be implemented, such as a proportion of the image area occupied by the strands from which it is considered that a sufficient number of strands have been detected. This proportion may be calculated as a function of the volume ratio of the fibers 21, 22 and the matrix 23 in the part 20. This proportion is typically of the order of 60%.

[0085] At the end of the process, the information obtained is recorded, for possible processing and / or digital reconstruction of the part 20. This information includes, for each strand, the position of the center of the strand, the dimensions of the characteristic shape corresponding to the section of the strand and its orientation.

[0086] The presented process therefore allows automatic identification of the strands in the part and obtaining the corresponding data, without direct intervention of an operator. The strong points of this process are therefore automation and reduction of the risk of errors, while ensuring high versatility of the analysis, thanks to the consideration of several parameters that can vary such as: the size of the strands, the shape of the strands as well as their orientations around their center.

Claims

1. Method for tomographic analysis of a composite part (20) comprising a matrix (23) and fibers (21, 22) embedded in the matrix (23), the method comprising the following steps: - acquiring at least one two-dimensional image of the part (20) by means of a tomographic device (1), characterized in that the method further comprises the steps: - generating a plurality of characteristic shapes (40) of the fibers (21, 22) and, optionally, for each characteristic shape (40), a convolution mask (50, 51, 52) comprising a plurality of copies of the characteristic shape (40), - calculating, for each of the characteristic shapes (40), a convolution product of the two-dimensional image with the characteristic shape (40) or with the corresponding convolution mask (50, 51, 52), and obtaining a convolution image, - detecting a position in the two-dimensional image corresponding to a global maximum overall the resulting convolution images for each of the characteristic shapes (40), and assignment of the fiber center (21, 22) to said position, - marking as processed a region of the image placed around the fiber center and corresponding to the characteristic shape (40) for which the maximum was obtained, and - iterating the steps of calculating, detecting and marking on the unprocessed regions of the image.

2. Method according to the preceding claim, wherein the method comprises a step of preprocessing the image, comprising at least one among a filtering of an average value of the image, and removal of edges from the image.

3. Method according to one of the preceding claims, wherein the acquisition step for the at least one two-dimensional image may comprise the acquisition of a three-dimensional image (30) of the part (20) by means of the tomographic device (1) and the implementation of at least one planar section in the three-dimensional image (30) to obtain each two-dimensional image.

4. Method according to one of the preceding claims, wherein the characteristic shapes (40) of the fibers (21, 22) are ellipses.

5. Method according to one of the preceding claims, wherein among the generated characteristic shapes (40), at least two characteristic shapes (40) have different dimensions from each other.

6. Method according to one of the preceding claims wherein among the generated characteristic shapes (40), at least two characteristic shapes (40) have different inclinations from each other relative to a reference direction of the two-dimensional image.

7. Method according to one of the preceding claims, wherein among the convolution masks (50, 51, 52) generated, at least two convolution masks (50, 52) comprise characteristic shapes (40) arranged with different separations between said two convolution masks (50, 52).

8. Method according to one of the preceding claims, wherein the part (20) is a part of a turbomachine housing, or a vane of a compressor rotor, stator or fan of the turbomachine.

9. Computer program product comprising instructions allowing implementation of the method according to one of the preceding claims when it is executed in a computer linked to a tomographic device.