Methods for automatically reconfiguring the reinforcing architecture of composite materials

Abstract

One aspect of the present invention relates to a method for automatically reconstructing the reinforcing structure of a composite material along a reinforcing axis, the method comprising the steps of: - acquiring each image along a cross section perpendicular to the reinforcing axis; - for each acquired image, detecting the centroid and / or circumference of each cross section of the reinforcing line using an artificial neural network; - for at least one acquired image, assigning a label corresponding to the reinforcing line to each detected centroid or each detected circumference; - for each other acquired image, assigning a label corresponding to the centroid to each detected centroid and / or each detected circumference in the acquired reference image; - reconstructing the structure of each reinforcing line from each detected centroid and / or from each detected circumference, each detected centroid and / or each detected circumference having the label of the reinforcing line and a position on the reinforcing axis associated with the acquired image, on which the centroid and / or circumference has been detected.

Description

Technical Field

[0001] The technical field of this invention is the technical field of composite materials, and more particularly the technical field of methods for reconstructing the architecture of reinforcing members of composite materials.

[0002] This invention relates to a method for reconstructing the architecture of a reinforcing member in a composite material, and more particularly to a method for automatically reconstructing the architecture of a reinforcing member in a composite material. The invention also relates to a computer program product and a recording medium that enable the implementation of the reconstruction method. Background Technology

[0003] In the field of composite materials, reconstructing the architecture of the reinforcing elements of a composite material by identifying each reinforcing thread (also called a strand) within the volume of the composite material is an excellent way to obtain the actual internal geometry of the material. Specifically, reconstructing the neutral fibers of each reinforcing thread—that is, the threads connecting the centroids along the entire length of each reinforcing thread—allows for obtaining the weave topology, enabling, for example, the searching of weave anomalies, and reconstructing the envelope of each reinforcing thread allows for obtaining the morphology of the fabric necessary for calculations of local thermophysical and thermomechanical properties. Identifying each reinforcing thread also allows for differentiation of each reinforcing thread, which may have different dimensions, i.e., include different numbers of fibers.

[0004] This allows for the detection of defective parts once they are manufactured, or the quantification of wear on parts made of composite materials.

[0005] Currently, reconstruction is performed using tomographic images, on which the operator has manually or the image processing algorithm has automatically identified the centroid of the cross section of each strand / line of the reinforcement.

[0006] In both cases, the operation is extremely time-consuming; a simple LEAP... TM The blades consist of thousands of carbon wires and are therefore subject to a lot of error, because it is sometimes impossible (even with the naked eye) to distinguish the wires of the reinforcing member from one another due to their high density.

[0007] Therefore, there is a need for a method for reconstructing the architecture of composite material reinforcements that is automated and has a reduced risk of errors. Summary of the Invention

[0008] By enabling the automatic acquisition of an accurate numerical model of a composite material reinforcement, the present invention provides a solution to the aforementioned problems.

[0009] A first aspect of the invention relates to a method for automatically reconstructing the architecture of a reinforcement member of a composite material along a reinforcement axis, the reinforcement member comprising a plurality of reinforcement lines arranged along the reinforcement axis, the method comprising the following steps:

[0010] - Acquire multiple images of the composite material reinforcement along a cross section parallel to the other acquired images and not confused with the cross section of the other acquired images, with each acquired image associated with a position on the reinforcement axis perpendicular to the cross section;

[0011] - For each acquired image, a first artificial neural network trained with a first training database is used to detect the centroid of each cross section of the reinforcement line present in the acquired image and / or a second artificial neural network trained with a second training database is used to detect the circumference of each cross section of the reinforcement line present in the acquired image;

[0012] - For at least one acquired image selected from the acquired images, referred to as the acquired reference image, a label corresponding to the reinforcement line is assigned to each detected centroid or each detected circumference in the acquired reference image;

[0013] - For each acquired image that is not selected as the acquired reference image, assign a label to the corresponding centroid or corresponding circle in the acquired reference image for each detected centroid and / or each detected circle in the acquired image.

[0014] - From each detected centroid and / or from each detected circumference, reconstruct the architecture of each reinforcement line arranged along the reinforcement axis, each detected centroid and / or each detected circumference having a label for the reinforcement line and a position on the reinforcement axis associated with the acquired image, on which the centroid and / or circumference have been detected.

[0015] As a result of this invention, a first artificial neural network enables the automatic detection of the centroid of each cross-section of the reinforcement line present in the acquired image, and / or a second artificial neural network enables the automatic detection of the circumference of each cross-section of the reinforcement line present in the acquired image. Labels allow each detected centroid and / or each detected circumference to be connected to the reinforcement line, thereby enabling the automatic reconstruction of its trajectory and / or its envelope.

[0016] By comparing the reconstructed trajectory of the reinforcing thread with the corresponding trajectory in the component's weave pattern, weave anomalies present in the component's reinforcement can be detected once the component is manufactured. Therefore, this operation allows for the automatic extraction of the fabric's topology and its comparison with a target topology.

[0017] The shape of the reinforcing wire's envelope allows it to possess the local geometric information necessary for assessing its mechanical properties. In fact, knowing the number of fibers constituting the wire and the wire's volume, the volume ratio of the fibers locally contained within can be determined, directly affecting its thermomechanical properties. Therefore, if the same composite material subjected to different external stresses is reconstructed several times, the surface deformation experienced by the wire can be known based on the external conditions applied, thus providing data-driven patterns that reveal its mechanical behavior.

[0018] In addition to the features already mentioned in the preceding paragraphs, the method according to the first aspect of the invention may have one or more additional features, either individually or in any technically permissible combination, of the following.

[0019] According to alternative implementations, the first artificial neural network and / or the second artificial neural network is a multilayer perceptron or a convolutional artificial neural network.

[0020] According to an alternative implementation compatible with the foregoing alternative implementation, each of the first and second training databases includes multiple images of at least one composite training material, each image being acquired along a cross-section perpendicular to the reinforcing axis of the composite training material.

[0021] Therefore, the first and second artificial neural networks are trained on images similar to the acquired images, and on the acquired images, the first and second artificial neural networks perform their detections.

[0022] According to an alternative implementation compatible with the foregoing alternative implementation, the step of using a clustering algorithm for allocation is included.

[0023] According to an alternative implementation compatible with the foregoing alternative implementation, images are acquired using a 3D imaging system with a resolution higher than a threshold.

[0024] According to a sub-alternative implementation of the aforementioned alternative implementation, the threshold is 150 μm for images acquired for centroid detection, or 40 μm for images acquired for circumference detection.

[0025] According to a sub-alternative implementation of the aforementioned alternative implementation that is compatible with the aforementioned sub-alternative implementation, images are acquired by synchrotron radiation imaging, by transmission electron microscopy imaging, or by X-ray tomography.

[0026] A second aspect of the invention relates to a method for automatically reconfiguring the architecture of a reinforcement in a composite material, the method comprising:

[0027] - According to a first aspect of the invention, for each reinforcing axis of the composite material, the steps of a method for automatically reconstructing the structure of the reinforcing member of the composite material along the reinforcing axis;

[0028] - The steps of reconstructing the structure of the reinforcing member of the composite material from the structure of each reinforcing line along each reinforcing axis.

[0029] Therefore, by reconstructing the trajectory and / or envelope of each reinforcing line, the reinforcing member of the composite material can be reconstructed, and thus the braided topology and / or the thermomechanical properties of the entire reinforcing member can be obtained.

[0030] A third aspect of the invention relates to a computer program product comprising instructions that, when executed by a computer, cause the computer to perform the steps of the method according to the first and / or second aspects of the invention.

[0031] A fourth aspect of the invention relates to a computer-readable recording medium comprising instructions that, when executed by the computer, cause the computer to perform the steps of the methods according to the first and / or second aspects of the invention.

[0032] A better understanding of the invention and its various applications will be gained by reading the following description and viewing the accompanying drawings. Attached Figure Description

[0033] These figures are presented for informational purposes and are in no way intended to limit the invention.

[0034] Figure 1 shows a three-dimensional reconstruction of the architecture of the composite reinforcement from an X-ray tomography image.

[0035] Figure 2 shows a schematic representation of a portion of the composite material reinforcement on the left and a portion of the separated reinforcement line on the right.

[0036] Figure 3 shows an X-ray tomographic image of the composite reinforcement obtained during the first step of the method according to the first aspect of the invention.

[0037] Figure 4 shows an X-ray tomographic image of the composite reinforcement obtained during the first step of the method according to the first aspect of the invention, indicating the center of gravity detected during the second step of the method according to the first aspect of the invention.

[0038] Figure 5 shows an X-ray tomographic image of the composite reinforcement obtained during the first step of the method according to the first aspect of the invention, indicating the circumference detected during the third step of the method according to the first aspect of the invention.

[0039] Figure 6 shows a graphical representation of a labeled reference image during the fourth step of the method according to the first aspect of the invention, and a graphical representation of an image showing a cross-section of a reinforcing line with a label being applied during the fifth step of the method according to the first aspect of the invention.

[0040] Figure 7 shows the position of each detected center of gravity in a reference frame defined by the reinforcing axis and an axis parallel to the cross section during the second step of the method according to the first aspect of the invention.

[0041] Figure 8 is a block diagram illustrating the steps of a method according to a first aspect of the present invention.

[0042] Figure 9 is a block diagram illustrating the steps of a method according to a second aspect of the present invention. Detailed Implementation

[0043] Unless otherwise specified, the same elements appearing in different figures have unique figure reference numerals.

[0044] The first aspect of the invention relates to a method for automatically reconstructing the structure of a reinforcing member of a composite material along a reinforcing axis.

[0045] A composite material is a component that includes at least one frame (called a reinforcing member) that includes reinforcing wires (also called strands) and an adhesive (called a matrix) that is typically made of a plastic material.

[0046] The reinforcing thread consists of multiple reinforcing fibers.

[0047] Reinforcing lines are arranged along at least one axis (referred to as the reinforcing axis).

[0048] A reinforcement is, for example, a stack of reinforced folds including reinforcing lines.

[0049] Figure 1 shows a three-dimensional reconstruction of the architecture of the composite reinforcement 300 from an X-ray tomography image.

[0050] In Figure 1, the upper reinforcing fold 301 is composed of reinforcing lines 302 arranged along axis Y and reinforcing lines 302 arranged along axis X. Therefore, axes X and Y are the reinforcing axes of the reinforcing member 300 shown in Figure 1.

[0051] In Figure 1, the reinforcing axes X and Y are substantially perpendicular, but the reinforcing fold 301 may include reinforcing lines 302 arranged along reinforcing axes forming an angle other than 90°. For example, the reinforcing axes X and Y may form an angle of 45°.

[0052] Figure 2 shows a schematic representation of a portion of the composite material reinforcement 300.

[0053] In Figure 2, a portion of the reinforcing member 300 includes three reinforcing folds 301-1, 301-2 and 301-3. The first reinforcing fold 301-1 includes four reinforcing lines 302-10, 302-11, 302-12, and 302-13; the second reinforcing fold 301-2 includes four reinforcing lines 302-20, 302-21, 302-22, and 302-23; and the third reinforcing fold 301-3 includes four reinforcing lines 302-30, 302-31, 302-32, and 302-33. Each reinforcing line 302-10, 302-11, 302-12, 302-13, 302-20, 302-21, 302-22, 302-23, 302-30, 302-31, 302-32, and 302-33 is arranged along the reinforcing axis Y.

[0054] The reinforcing folds 301 of the same reinforcing member 300 can have different reinforcing axes X and Y. For example, the reinforcing member 300 includes two types of reinforcing folds 301, each of the first type and the second type of reinforcing folds 301 including two reinforcing axes X and Y forming a 45° angle, and overlapping in such a way that the reinforcing axes X and Y of the first type and the reinforcing axes X and Y of the second type form a 45° angle, then the reinforcing member 300 includes four different reinforcing axes.

[0055] The term "reconstructing the architecture of a composite reinforcement" refers to obtaining a numerical model of the architecture of a composite reinforcement, which can then be used for numerical characterization of the composite material, such as the thermophysical and / or thermomechanical properties at each point of the reinforcement or at the location of weaving anomalies.

[0056] The method according to the first aspect of the invention makes it possible to obtain a reconstruction of the structure of the composite reinforcement 300 along the reinforcing axis, for example, along axis X or along axis Y in the case of FIG1, and along axis Y in the case of FIG2.

[0057] In the remainder of this specification, the method according to the first aspect of the invention should be applied to the reinforcing axis Y, that is, the structure of the composite material reinforcement 300 should be determined along the reinforcing axis Y.

[0058] Figure 8 is a block diagram illustrating the steps of a method 100 according to a first aspect of the present invention.

[0059] The first step 101 of method 100 includes acquiring multiple images of the composite material reinforcement 300.

[0060] For example, images can be acquired using a 3D imaging system with a resolution above a threshold (e.g., synchrotron radiation imaging or transmission electron microscopy imaging system) or by X-ray tomography.

[0061] Acquire each image along a section perpendicular to the reinforcing axis Y.

[0062] Figure 2 shows a graphical representation of the separate reinforcing lines 302-22 arranged along the reinforcing axis Y and intersecting with the section 3031 perpendicular to the reinforcing axis Y.

[0063] As shown in Figure 2, the image obtained along section 3031 includes the section of reinforcement line 302. This section can be defined by its centroid 3021 and its circumference 3022.

[0064] For example, the resolution threshold of the 3D imaging system is 150 μm so that the centroid 3021 of the reinforcement line 302 can be detected in the acquired image 303, and 40 μm so that the circumference 3022 of the reinforcement line 302 can be detected in the acquired image 303.

[0065] Figure 3 shows an image 303 acquired by X-ray computed tomography during the first step 101 of method 100.

[0066] Image 303 is acquired along section 3031, which includes axes X and Z and is perpendicular to the reinforcing axis Y. Image 303 includes sections of multiple reinforcing lines 302.

[0067] The cross sections 3031 of the image 303 acquired during the first step 101 are parallel to each other and not confused; that is, the cross sections 3031 are spaced apart between them along the reinforcing axis Y. Therefore, each cross section 3031 can be associated with a position on the reinforcing axis Y.

[0068] For example, if three images 303 perpendicular to the reinforcing axis Y are acquired along section 3031 (each section is spaced 1 mm apart), then, for example, the first image 303 is associated with a second image 303 at position 0 mm on the reinforcing axis Y, a third image 303 at position 1 mm on the reinforcing axis Y, and a third image 303 at position 2 mm on the reinforcing axis Y.

[0069] Section 3031 may or may not be spaced at the same distance along the reinforcing axis Y.

[0070] The second step 102 of method 100 includes using a first artificial neural network trained with a first training database to detect the centroid 3021 of each section of the reinforcement line 302 in each of the images 303 acquired above.

[0071] An artificial neural network comprises at least one layer of artificial neurons, with each layer containing at least one artificial neuron. The artificial neurons in an artificial neural network are connected together by synapses, and each synapse is assigned a synaptic coefficient.

[0072] This training enables the artificial neural network to be trained for a predetermined task by updating the synaptic coefficients in a way that minimizes the error between the output data provided by the artificial neural network and the actual output data. In other words, it determines what the artificial neural network should provide as output to complete a predetermined task on specific input data.

[0073] Therefore, the training database includes input data, with each input data being associated with the actual output data.

[0074] The first artificial neural network has the function of detecting the centroid 3021 of each cross section of the reinforcing line 302 in the image 303 acquired along a cross section perpendicular to the reinforcing axis Y.

[0075] Therefore, the first training database includes images of the same or several composite training materials, each image is acquired along a cross section perpendicular to the reinforcement axis of the composite training material, and includes data on the position of the centroid 3021 of each cross section of the reinforcement line 302 in each image 303.

[0076] One or more composite training materials may be the same as or different from the composite material desired for use in the architecture of the reconstructed reinforcement 300.

[0077] According to the method 100 of the first aspect of the present invention, the structure of the composite reinforcement 300 can be reconstructed before or after assembly with the matrix.

[0078] Therefore, the images in the first training database can be reinforced images 300 of composite materials before assembly with their matrices and / or reinforced images 300 of composite materials after assembly with their matrices.

[0079] Figure 4 shows an image 303 acquired by X-ray computed tomography during the first step 101 of method 100. The centroid 3021 of each section of the reinforcing line 302 is shown in it.

[0080] As shown in Figure 3, image 303 of Figure 4 is obtained along section 3031, which includes axes X and Z perpendicular to the reinforcing axis Y.

[0081] Each centroid 3021 detected in image 303 is associated with a position in section 3031, i.e., in a reference frame defined by axes X and Z.

[0082] The third step 103 of method 100 includes using a second artificial neural network trained with a second training database to detect the circumference 3022 of each cross section of the reinforcement line 302 in each of the images 303 acquired above.

[0083] The second artificial neural network has the function of detecting the circumference 3022 of each section of the reinforcing line 302 in the image 303 acquired along the section 3031 perpendicular to the reinforcing axis Y.

[0084] Therefore, the second training database includes images of one or more composite training materials, each image is acquired along a cross section perpendicular to the reinforcement axis of the composite training material, and includes data on the location of different points on the circumference 3022 of each cross section of the reinforcement line 302 in each image 303.

[0085] The images in the first training database can be the same as the images in the second training database.

[0086] Data on the position of the centroid 3021 included in the first training database and data on the position of points on the circumference 3022 included in the second training database can be obtained manually or through image processing algorithms (e.g., using mathematical morphology algorithms).

[0087] For example, the first and second artificial neural networks are multilayer perceptrons or convolutional artificial neural networks.

[0088] Figure 5 shows image 303 acquired by X-ray computed tomography during the first step 101 of method 100. The circumference 3022 of each section of the reinforcing line 302 is shown in image 303.

[0089] As shown in Figures 3 and 4, image 303 of Figure 5 is obtained along section 3031, which includes axes X and Z perpendicular to the reinforcing axis Y.

[0090] Each point of the detected circumference 3022 is associated with a position in section 3031, i.e., in a reference frame defined by axes X and Z.

[0091] The second step 102 or the third step 103 of method 100 can be optional. That is, the architecture of the composite material reinforcement 300 can be reconstructed by performing only the second step 102, or by performing only the third step 103, or by performing both the second step 102 and the third step 103.

[0092] The fourth step 104 of method 100 includes, if the second step 102 above has been performed, assigning a label corresponding to the reinforcement line 302 to each detected centroid 3021 in one or more images 303 selected above as reference images, or if the third step 103 above has been performed, assigning a label corresponding to the reinforcement line 302 to each detected circumference 3022 in one or more images 303 selected above as reference images.

[0093] In other words, during the fourth step 104, if only the second step 102 is performed, or if both the second step 102 and the third step 103 are performed, a label is assigned to each centroid 3021 in the reference image; or if only the third step 103 is performed, or if both the second step 102 and the third step 103 are performed, a label is assigned to each circumference 3022 in the reference image.

[0094] The term "label assigned to an element" refers to a label assigned to an element so that it can be identified and distinguished from other elements.

[0095] For example, a reference image is selected for every N acquired images 303. For example, if there are 20 acquired images 303 and it is desired to select a reference image for every 5 images, then, for example, the first acquired image 303, the sixth acquired image 303, the eleventh acquired image 303, and the sixteenth acquired image 303 can be selected.

[0096] Figure 6 shows a schematic representation of the acquired reference image 303'. The acquired reference image 303' includes three sections of the reinforcing line 302, and therefore includes three centroids 3021. Each centroid 3021 is associated with a label L1, L2, or L3. Thus, the first centroid 3021, positioned towards the upper left corner, corresponds to label L1; the second centroid 3021, positioned towards the middle right side, corresponds to label L2; and the third centroid 3021, positioned towards the middle bottom, corresponds to label L3.

[0097] The fifth step 105 of method 100 includes: if the second step 102 above has been performed, assigning each centroid 3021 present in the acquired image 303 to each acquired image 303 that was not selected as an acquired reference image 303', and / or if the third step 103 above has been performed, assigning each circumference 3022 present in the acquired image 303 to each acquired image 303 that was not selected as an acquired reference image 303', wherein the centroid 3021 and / or the corresponding circumference 3022 in the acquired reference image 303' have been labeled in the fourth step 104.

[0098] In other words, if a label is assigned to the centroid 3021 in the reference image during the fourth step 104, the fifth step 105 includes assigning a label to each acquired image 303 that was not selected as an acquired reference image 303', and assigning a label to each centroid 3021 present in the acquired image 303 and the corresponding centroid 3021 in the acquired reference image. If the third step 103 was also performed, the fifth step 105 includes assigning a label to each acquired image 303 that was not selected as an acquired reference image 303', and assigning a label to each circle 3022 present in the acquired image 303 and the corresponding centroid 3021 in the acquired reference image.

[0099] If a label is assigned to a circumference 3022 in the reference image during the fourth step 104, then the fifth step 105 includes assigning a label to each acquired image 303 that was not selected as an acquired reference image 303', to each circumference 3022 present in the acquired image 303 and the corresponding circumference 3022 in the acquired reference image, and if the second step 102 was also assigned, the fifth step 105 includes assigning a label to each acquired image 303 that was not selected as an acquired reference image 303', to each centroid 3021 present in the acquired image 303 and the corresponding circumference 3022 in the acquired reference image.

[0100] For example, in the case of labeling the centroid 3021, if six centroids 3021 are labeled in the acquired reference image 303' immediately preceding the image 303 to which the fourth step 104 is performed, then each of the six centroids 3021 detected in the acquired image 303 will be associated with one of the labeled centroids 3021 in the acquired reference image 303'.

[0101] In Figure 6, the acquired image 303 includes three centroids 3021: a first centroid 3021 located at the upper left corner, a second centroid 3021 located towards the right center, and a third centroid 3021 located towards the bottom center. By comparing the acquired image 303 with the immediately preceding acquired reference image 303', for example, by selecting the label corresponding to the centroid 3021 of the acquired reference image 303' that has the smallest distance from the detected centroid 3021 in the acquired image 303, the centroid 3021 located at the upper left corner can be associated with label L1, the centroid 3021 located towards the bottom center can be associated with label L3, and the centroid 3021 located towards the right center can be associated with label L2.

[0102] For example, step 105 is performed using a clustering algorithm (e.g., the DBSCAN algorithm).

[0103] At the end of step 5 105, the centroid 3021 and / or circumference 3022 of each acquired image 303 are marked. Centroids 3021 and / or circumferences 3022 with the same label correspond to the same reinforcing line 302.

[0104] The sixth step 106 of method 100 includes reconstructing the architecture of each reinforcing line 302 arranged along the reinforcing axis Y.

[0105] In the case of the second step 102 of method 100, the architecture of each reinforcement line 302 is reconstructed from the position of each centroid 3021 of the label with reinforcement line 302 in its cross section 3031, and from the position on the reinforcement axis Y associated with the cross section 3031 and therefore associated with the acquired image 303 in which the centroid 3021 is detected.

[0106] Returning to a simplified embodiment of acquiring three images 303 along a plane at 1mm intervals, the trajectory of the reinforcing line 302 can be approximated if the position of the centroid 3021 of the reinforcing line 302 in each of the images 303 acquired along a section 3031 perpendicular to the reinforcing axis Y is known.

[0107] For example, if the centroid 3021 of the reinforcing line 302 is located at coordinate point (X1, Z1) in the reference system defined by axes X and Z in the first image 303, then at coordinate point (X2, Z2) in the second image 303, and then at coordinate point (X3, Z3) in the third image 303, the reinforcing line 302 is known to have a trajectory that passes through point (X1, Z1) at position 0 mm on the reinforcing axis Y, passes through point (X2, Z2) at position 1 mm, and passes through point (X3, Z3) at position 2 mm.

[0108] Therefore, by minimizing the interval between the cross sections 3031 of the different acquired images 303, the accuracy of the approximation of the trajectory of the reinforcing line 302 is improved.

[0109] Figure 7 shows the position of each centroid 3021 detected during the second step 102 of method 100 in a reference frame defined by the reinforcing axis Y and the axis X parallel to section 3031.

[0110] Returning to the foregoing embodiment, this returns to placing the three centers of gravity 3021 in a reference frame defined by axes X and Y, i.e., placing points X1, X2, and X3 on the reinforcing axis Y according to the distances associated with the acquired image 303.

[0111] In Figure 7, for a given value along axis X, up to four centroids 3021 are shown, which means there are up to four different reinforcing lines 302.

[0112] By adding information about the label associated with the center of gravity 3021, as shown in Figure 7, the trajectories of different reinforcing lines 302 can be distinguished for the reinforcing line 302 associated with label L1.

[0113] In the case of the third step 103 of method 100, the architecture of each reinforcing line 302 is reconstructed from the position of each point of the circumference 3022 of the label with reinforcing line 302 in its cross section 3031, and from the position on the reinforcing axis Y associated with the cross section 3031 and thus from the acquired image 303 in which the center of gravity 3021 is detected.

[0114] If steps 102 and 103 are performed, Figure 7 can be supplemented by increasing the size of the reinforcing line 302 along axis X.

[0115] A second aspect of the invention relates to a method for automatically reconstructing the complete architecture of a reinforcing member 300 of a composite material.

[0116] Figure 9 is a block diagram illustrating the steps of a method 200 according to a second aspect of the present invention.

[0117] For each reinforcing axis X, Y of the reinforcing member 300, method 200 includes the steps of method 100 according to the first aspect of the invention.

[0118] If, as in the embodiments described above, the reinforcement 300 includes four reinforcing axes, then for each of the four reinforcing axes, the method according to the second aspect of the invention includes the steps of the method 100 according to the first aspect of the invention.

[0119] In Figure 9, the steps of method 100 are performed twice, which means that the reinforcement 100 of the desired reconstructed architecture has two reinforcement axes X and Y.

[0120] At the end of these steps, the architecture of each reinforcing line 302 is known because each reinforcing line 302 is arranged along any reinforcing axis X, Y.

[0121] Then, the final step 201 involves reconstructing the complete architecture of the composite material reinforcement 300 from the architecture of each reinforcing line 302 of the reinforcement 300.

[0122] For example, in order to reconstruct the 3D architecture of the stiffener 300 in Figure 1, it is sufficient to place each stiffener line 302 in a reference frame defined by the axes X, Y and Z.

Claims

1. Method (100) for automatically reconstructing the architecture of a reinforcement (300) of a composite material along a reinforcement axis X or Y, the reinforcement (300) comprising a plurality of reinforcement lines (302) arranged along a reinforcement axis X or Y, the method (100) being characterized in that it comprises the following steps: - a step (101) of acquisition of a plurality of images (303) of the reinforcement (300) of a composite material, each image (303) being acquired along a section (3031) parallel to the sections (3031) of the other acquired images (303) and not confused with the sections (3031) of the other acquired images (303), each acquired image (303) being associated with a position on the reinforcement axis X or Y perpendicular to the section (3031); - a step (102) of detection, for each acquired image (303), of the center of gravity (3021) of each section of reinforcement line (302) present in the acquired image (303) using a first artificial neural network trained with a first training database and / or - a step (103) of detection, using a second artificial neural network trained with a second training database, of the circumference (3022) of each section of reinforcement line (302) present in the acquired image (303); - a step (104) of assignment, for at least one acquired image chosen from the acquired images (303), called acquired reference image (303'), to each detected center of gravity (3021) or to each detected circumference (3022) in the acquired reference image (303'), of a label corresponding to a reinforcement line (302); - a step (105) of assignment, for each acquired image (303) that is not chosen as an acquired reference image (303'), to each detected center of gravity (3021) and / or to each detected circumference (3022) in the acquired image (303), of the label of the corresponding center of gravity (3021) or of the corresponding circumference (3022) in the acquired reference image (303'); - a step (106) of reconstruction, from each detected center of gravity (3021) and / or from each detected circumference (3022), of the architecture of each reinforcement line (302) arranged along a reinforcement axis X or Y, said each detected center of gravity (3021) and / or said each detected circumference (3022) having the label of a reinforcement line (302) and a position on the reinforcement axis X or Y associated with the acquired image (303) on which the center of gravity (3021) and / or the circumference (3022) has been detected.

2. The method (100) according to claim 1, characterized in that The first and second artificial neural networks are multilayer perceptrons and / or convolutional artificial neural networks.

3. The method (100) according to claim 1, characterized in that The first training database and the second training database each comprise a plurality of images of at least one composite training material, each image being acquired along a section perpendicular to the reinforcement axis of the composite training material.

4. The method (100) of claim 1, characterized by The assignment step (105) is performed using a clustering algorithm.

5. The method (100) of claim 1, characterized by The images (303) are acquired by means of a 3D imaging system having a resolution higher than a threshold value.

6. The method (100) according to claim 5, characterized by The threshold is 150 pm for the image (303) acquired for the detection of the barycenter (3021) or the threshold is 40 pm for the image (303) acquired for the detection of the circumference (3022).

7. The method (100) according to claim 5, characterized by The image (303) is acquired by synchrotron radiation imaging, by transmission electron microscopy or by X-ray tomography.

8. Method (200) for automatically reconfiguring the architecture of a reinforcement (300) of a composite material, characterized in that, The method comprises: - the steps (101), (102), (103), (104), (105) and (106) of the method (100) for automatically reconstructing the architecture of the reinforcement (300) of the composite material along each reinforcement axis X and Y according to any one of claims 1 to 7, - the step of reconstructing (201) the architecture of the reinforcement (300) of the composite material from the architecture of each reinforcement line (302) along each reinforcement axis X and Y.

9. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the steps of the method (100) or method (200) according to any one of claims 1 to 8.

10. A record medium readable by a computer, the record medium comprising instructions which, when executed by a computer, cause the computer to carry out the steps of the method (100) or method (200) according to any one of claims 1 to 8.

Images