PROCEDURE OF DETERMINATION OF A REPRESENTATION OF A SKELETTE OF A BULL OF A WOVEN COMPOSITE MATERIAL

The use of representation learning techniques addresses the inefficiencies in existing methods by synthesizing database information to enhance the spatial representation of woven composite material strands, improving speed, reproducibility, and accuracy.

FR3148838A1Pending Publication Date: 2024-11-22SAFRAN SA +3
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
FR2023004811
Authority / Receiving Office
FR · FR
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-15
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing methods for extracting the textile model of woven composite materials are time-consuming, costly, and lack reproducibility and reliability due to manual intervention and error propagation in automatic methods, which are often specific to particular strand crossing patterns.

Method used

A method using representation learning techniques, such as autoencoders or recurrent neural networks, to synthesize information from a database of strand structures, reducing the need for manual identification points and improving the spatial representation of strand skeletons.

Benefits of technology

The method significantly reduces the number of manual steps required and enhances the reproducibility and accuracy of determining the spatial structure of woven composite material strands.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

METHOD FOR DETERMINING A REPRESENTATION OF A SKELETON OF A STRAND OF WOVEN COMPOSITE MATERIAL. A computer-implemented method (100) for determining a spatial representation of a skeleton of a strand of a portion of a part made of woven composite material, the strand extending in a predetermined direction, said method comprising: Obtaining (110), from a database comprising spatial representations of skeletons of strands extending in the predetermined direction, a latent space of the database, the database being described in the latent space by a set of latent variables; Receiving (120) a set of spatial coordinates of the strand; Determining (130) the spatial representation of the skeleton of the strand from the set of spatial coordinates obtained, said spatial representation of the skeleton of the strand being a function of the latent variables. Figure to be published with the abstract: Figure 4
Need to check novelty before this filing date? Find Prior Art

Description

Title of the invention: METHOD FOR DETERMINING A REPRESENTATION OF A SKELETON OF A STRAND OF A WOVEN COMPOSITE MATERIAL TECHNICAL FIELD OF THE INVENTION

[0001] The technical field of the invention is that of woven composite materials and in particular that of the digital modeling of these materials.

[0002] In particular, the invention relates to a method for determining a skeleton of a strand of a woven composite material using a machine learning model using a representation learning technique. The invention also relates to a device for determining a spatial structure of a skeleton of a woven composite material. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0003] A woven composite material is an assembly comprising at least one woven textile framework called reinforcement and a binder called matrix. The reinforcement is the frame that takes up most of the mechanical forces. The matrix has the purpose of transmitting the mechanical forces to the reinforcement, ensuring its protection and its arrangement within the part.

[0004] The reinforcement comprises strands (also called "threads") woven using a loom, following a theoretical weaving topology. The weaving topology is defined according to at least two orientations, also called reinforcing axes, each strand comprising a plurality of fibers, often carbon or glass. The reinforcing axes are conventionally called "warp" and "weft". In general, the warp corresponds to the main weaving direction, and the weft corresponds to the transverse direction, orthogonal to the warp. The way in which the strands are interwoven (i.e. the pattern according to which the strands are woven) is conventionally called "weave".

[0005] Conventionally, a textile model comprises a parametric description of each strand within a weave. Here, the term "parametric description of a strand" means a description of a strand as a solid object, characterized by a neutral fiber and cross-sections associated with at least certain points of the neutral fiber.

[0006] The term "neutral fiber" or "skeleton" is understood to mean a curve describing the central path of the strand in space. Cross-sections are two-dimensional curves defined on planes orthogonal to the tangent to the neutral fiber. An example of a parametric description of a strand is shown in [Fig.l]. The neutral fiber is here represented by the ordered series of points connected by the solid line. The ellipses represent the cross-sections of the parametric representation strand metric. Textile modeling using the strand as the most basic unit is also called mesoscale textile modeling.

[0007] There are methods for determining the textile model of a part (or a portion of a part) made of composite material from a 3D tomographic image of this part. An example of a tomographic image of such a woven material is shown in [Fig.2] (adapted from the article “Descriptive modeling of textiles using FE simulations and deep learning”, by A. Mendoza, R. Trullo and Y. Wielhorski published in the journal “Composites Science and Technology” in 2021). One of these methods is based on the manual extraction, by expert operators, of neutral fibers and cross-sections.

[0008] Such a manual approach to textile model extraction begins with the manual identification of the neutral fibers of all the strands present within a region of interest of the woven composite material. It is important to note that this manual identification step is very long. Then, one or more cross sections are assigned to the neutral fiber so as to obtain a solid object for each strand. In general, a very simple parametric cross section (e.g. an ellipse) is assigned to each orientation (warp and weft). This step therefore does not require intervention by the expert operator for each strand or for each important point (e.g. a point associated with a change in curvature) of the strand. Thus, it is the step of manual identification of the neutral fibers which is essential in the manual approach to textile model extraction.

[0009] The procedure for manually locating a strand j in the region of interest is detailed in Figure 3. The first step involves identifying the planes of interest for strand j, represented by the vertical dashed lines. This locating is performed on a two-dimensional image or two-dimensional section representing the strand in its length, such as that shown on the left. In this case, it is a section along the xz plane, on which the coordinates x) along the x axis of the centers of the neutral fiber at the planes of interest are determined, each value x) corresponding to the coordinate along the x axis of the center of the neutral fiber associated with strand j at plane i. The other two coordinates of the centers of the neutral fiber at the planes of interest, namely _A, are then identified on an xy section corresponding to the V j- j) plane of interest i (illustrated on the right in [Fig.3]). These two steps make it possible to identify all the points of the neutral fiber associated with the strand and corresponding to the planes of interest. The planes of interest are generally fixed by the operator according to the crossings (of the strands in the orthogonal direction) that the strand presents in its length or by its inflection points.

[0010] Then, interpolation functions (for example splines or functions of radial basis) are used to connect the identified points, so as to obtain the neutral fiber associated with the strand. For each strand, a respective 3D curve is thus obtained. However, a comparison between the curve obtained and several two-dimensional views is necessary to verify the quality of the modeling obtained. If the neutral fiber thus obtained does not represent the strand of interest precisely enough, it is necessary to repeat the process by adding additional planes of interest.

[0011] This manual tracking process must be performed for each strand in the region of interest, making manual extraction of the textile model particularly time-consuming and costly. In addition, the results can be highly operator-dependent, limiting the reproducibility and reliability of manual extraction.

[0012] Recently, some methods for automatic or semi-automatic extraction of the textile model have been proposed. These methods are based on image processing techniques, for example to identify the regions of the volume belonging to each of the components of the woven material or to identify key points for the definition of the strands.

[0013] Most of these methods use the structure tensor to extract local orientation information. These signatures are then processed either by ad hoc algorithms built on the basis of business heuristics (see for example the scientific article “3D composite reinforcement meso FE analyses based on X-ray computed tomography” published in “Composite Structures” in 2015), or by algorithms based on statistical analysis of the data (see for example the scientific article “Quantification of the internal structure and automatic generation of voxel models of textile composites from X-ray computed tomography data” by I. Stramit et al., published in “Composites Part A: Applied Science and Manufacturing” in 2015). It has also been proposed to process these signatures by deep learning type algorithms (see for example “Descriptive modeling of textiles using FE simulations and deep learning” by A. Mendoza et al., published in “Composites Science and Technology” in 2021).

[0014] However, each of these methods is developed for a very specific strand crossing pattern and is not generally applicable. Furthermore, these methods are often not robust to errors (they tend to propagate center positioning errors from one cut to the next), and have no means of controlling such errors. There is therefore a need to develop tools to make the extraction of neutral fibers from strands of a woven composite material faster, more reproducible and more reliable. Summary of the invention

[0015] The invention provides a solution to the problems mentioned above by determining a spatial representation of a skeleton of a strand of a composite woven material from manually located points and a database comprising a multiplicity of strand structures. By applying a representation learning technique to the database, it is possible to synthesize the information contained in the database and use it to determine the structure of a new strand. By using the information extracted by analyzing the database, the number of points that need to be manually located to obtain a spatial representation of a skeleton of a strand is greatly reduced.

[0016] A first aspect of the invention thus relates to a method, implemented by computer, for determining a spatial representation of a skeleton of a strand of a portion of a part made of woven composite material, the strand extending in a predetermined direction x, said method comprising: • Obtaining, from a database comprising spatial representations of strand skeletons extending in the predetermined direction T, a latent space of the database, the database being described in the latent space by a set of latent variables; • Receive a set of spatial coordinates z(x) of the strand; • Determine the spatial representation of the strand skeleton from the set of spatial coordinates z(x) received, said spatial representation of the strand skeleton being a function of the latent variables.

[0017] By "strand" is meant an assembly of one or more bundles of fibers (also called wicks). Often the bundles of fibers are twisted together to improve the cohesion of the strand.

[0018] By "skeleton" is meant a parametric model of a strand comprising a series of spatial coordinates of centers of the strand. In other words, the skeleton (or "neutral fiber" of the strand) represents the "path" of the center of the strand in space inside the portion of the woven composite material part. By "portion of part" is meant all or part of the part.

[0019] By "spatial representation of a skeleton of a strand" is meant a representation describing the structure of the skeleton of the strand. The spatial representation may comprise, for example, a set of spatial coordinates in two dimensions or in three dimensions. Alternatively, the spatial representation may comprise one or more mathematical functions giving a spatial coordinate of the skeleton as a function of another spatial coordinate. For example, the spatial representation of a skeleton of a strand may have the form < — f(x)ouz — f(x, y).

[0020] By "latent space" is meant a space representing the elements of the database by means of their main characteristics. The latent space is described by latent variables which can be used to describe the elements of the database. data. The latent space is for example obtained by applying a representation learning technique to the database.

[0021] By "representation learning technique" is meant a machine learning method capable of automatically detecting characteristics making it possible to describe, often in a simplified form, the input database. For example, a representation learning technique may provide a representation of the input database using a reduced base. Examples of representation learning models are principal component analysis, autoencoders or recurrent neural networks.

[0022] The spatial representation of the strand skeleton is therefore obtained by minimizing a cost function determined from the distance between known points and their predictions provided by the model.

[0023] The above method makes it possible to use a simplified representation of the database to obtain, starting from the set of coordinates 4^) of the strand, a spatial representation of a skeleton of a strand. The representation thus obtained is a function of latent variables describing the database in the latent space.

[0024] The function expressing the skeleton can for example be found by minimizing a difference between the set of coordinates 4X) of the strand and a linear combination of latent variables.

[0025] The machine learning model thus trained is therefore generalized and capable of providing accurate predictions for any type of strand crossing pattern, provided that this pattern is sufficiently represented in the database.

[0026] In one or more embodiments, the latent space of the database is obtained using a representation learning technique chosen from a list comprising: principal component analysis, autoencoder, recurrent neural networks.

[0027] In one or more embodiments, the woven composite material part is an aeronautical part. For example, the aeronautical part is an engine blade.

[0028] In one or more embodiments: • Obtaining the latent space representative of the database includes a principal component analysis of said database, the latent variables being P principal components mi; • The spatial representation of the strand skeleton includes a set of points associated with respective coordinates ?(x), in which g(x) is equal to a linear combination of type yp a / \, with ai P coefficients ^^ / =1 f «A* / real.

[0029] In one or more embodiments, the coefficients at are determined in mi minimizing a gap between the received spatial coordinates and the z(Xq) values ​​of the linear combination evaluated for the positions x^.

[0030] In one or more embodiments, the minimized deviation is of the form

[0031] In one or more embodiments, obtaining the representative latent space of the database includes the application of an autoencoder type model to the database. Nq ^q=l I

[0032] In one or more embodiments: • Obtaining the latent space representative of the database includes applying an autoencoder type model to the database; • The latent variables of the database are of type y — f (y Y being a spatial representation of a skeleton belonging to the database.

[0033] In one or more embodiments, wherein obtaining the latent variables K from the database comprises determining two functions / , and r encode ^decode chosen so as to minimize the gap between X and r ((f / yY].

[0034] In one or more embodiments, obtaining the latent space representative of the database comprises applying a recurrent neural network type model.

[0035] A second aspect of the invention relates to a device for determining a spatial representation of a skeleton of a strand of a portion of a piece made of woven composite material, the strand extending in a predetermined direction x, said device including:

[0036] An input interface for receiving: • Latent variables describing a latent space of a database comprising spatial representations of strand skeletons extending in the predetermined x direction; • A set of spatial coordinates z(x) of the strand; • A circuit configured to determine the spatial representation of the strand skeleton from the set of spatial coordinates z(x), the spatial representation being a function of the latent variables.

[0037] The device according to the second aspect of the invention allows the determination of the spatial representation of the skeleton of the strand using the method according to the first aspect of the invention.

[0038] A third aspect of the invention relates to a device for determining a latent space of a database comprising spatial representations of strand skeletons extending in a predetermined x direction, said device comprising: • An input interface to receive the database; • A circuit for obtaining the latent space by applying a representation learning technique to the database; • An output interface to provide a set of latent variables representing the database in the latent space.

[0039] The above latent space determination device makes it possible, by applying a representation learning technique, to synthesize the information contained in the database by determining the main characteristics of the strand skeletons included in the database.

[0040] The representation learning technique is included in a set comprising: principal component analysis, auto-encoder, recurrent neural networks.

[0041] The latent variables provided by the device for determining the latent space according to the third aspect of the invention can be used by the device for determining the spatial structure of the strand according to the second aspect of the invention.

[0042] A fourth aspect of the invention relates to a computer program product comprising the instructions for implementing the method according to the first aspect of the invention when this program is executed by a processor.

[0043] The invention and its various applications will be better understood upon reading the following description and examining the accompanying figures. BRIEF DESCRIPTION OF THE FIGURES

[0044] Other characteristics and advantages of the invention will appear on reading the description, which can be read in conjunction with the figures. These figures are presented for information purposes only and in no way limit the invention.

[0045] [Fig.l] [Fig.l] schematically represents a parametric representation of a strand.

[0046] [Fig.2] [Fig.2] represents a tomographic image of a woven composite material.

[0047] [Fig.3] [Fig.3] schematically represents the method of manual extraction of a textile model according to the state of the art.

[0048] [Fig.4] [Fig.4] represents a flowchart of a method for determining a spatial representation of a skeleton of a strand according to the first aspect of the invention.

[0049] [Fig.5] [Fig.5] represents an example of a database comprising spatial representations of strand skeletons.

[0050] [Fig.6] [Fig.6] represents 6 principal components obtained from the base data from [Fig.6].

[0051] [Fig.7a] [Fig.7a] represents an example of implementation of the method according to the first aspect of the invention.

[0052] [Fig.7b] [Fig.7b] represents an example of implementation of the method according to the first aspect of the invention.

[0053] [Fig.8] [Fig.8] represents an example of a device for determining a skeleton of a strand according to the second aspect of the invention.

[0054] [Fig.9] [Fig.9] represents an example of a device for determining a latent space of a database according to the third aspect of the invention. DETAILED DESCRIPTION

[0055] Figures 1, 2 and 3 have been described in relation to the state of the art.

[0056] [Fig.4] represents a flowchart of the method 100 for determining a spatial representation of a skeleton of a strand according to the first aspect of the invention.

[0057] The method 100 comprises a step 110 of obtaining a latent space from a database comprising spatial representations of strand skeletons. The strand skeletons belonging to the database extend in the same direction as the strand whose skeleton is to be determined. For example, the direction may be a weft or warp direction of the woven composite material.

[0058] Step 110 therefore makes it possible to find the main characteristics of the skeletons forming the database and to represent them in the form of latent variables.

[0059] The method 100 according to the first aspect of the invention further comprises a step 120 of receiving a set of spatial coordinates z(x) of the strand whose spatial representation is to be determined.

[0060] The spatial representation of the strand is determined, during step 130, from the set of spatial coordinates z(x) and the latent variables determined during step 110. In particular, the spatial representation of the strand is obtained in the form of a function of the latent variables.

[0061] According to one embodiment, step 130 of determining the spatial representation of the skeleton of the strand comprises the minimization of a cost function taking into account the difference between the set of spatial coordinates of the strand and a function of the latent variables.

[0062] According to one embodiment of the method 100 according to the invention, the latent space of the database is determined by applying a representation learning technique.

[0063] Figure 5 illustrates a database comprising spatial representations z(x) of strand skeletons. In particular [Fig.5] represents 34 warp-direction strands, each strand being described by 433 points. The spatial representations of skeletons illustrated in [Fig.5] are two-dimensional and belong to the xz plane.

[0064] The spatial representations of skeletons shown in [Fig.5] result from manual identification followed by “spline” or “RBF” type interpolations.

[0065] According to an embodiment of the method 100 according to the first aspect of the invention, the representation learning technique applied to the database during step 110 is of the principal component analysis or PCA type. Figure 6 illustrates the first 6 principal components or eigenmodes m> obtained by applying PCA to the database illustrated in [Fig.5].

[0066] The 6 principal components, also called eigenmodes or principal modes, make it possible to describe 99% of the variance of the data illustrated in [Fig.5]. In other words, a linear combination of the 6 modes illustrated in [Fig.6] makes it possible to describe a skeleton of the database with an error less than 1% of the data variance.

[0067] In this case, the latent space of the database therefore comprises the first 6 principal modes mi of a principal component analysis of the database.

[0068] In other words, when the skeleton of the strand to be determined is evaluated on a set of Nq question points (xq, zq) with qg [1, NJ, the spatial representation of the strand determined using the method 100 according to the first aspect of the invention is in the form g ( x ) = ^ai' 'cs P°^s U' being determined to approximate the question points.

[0069] A question point is understood to mean a point on the skeleton of the strand which has been determined, for example, manually. Determining the complete skeleton includes reconstructing a curve passing through these points (or as close as possible to these points, as detailed below) of interrogation, which is representative of a strand skeleton.

[0070] According to one embodiment, the weights ai are determined using the method of least squares minimizing the deviation, [ ^p 2, P being the number of principal components.

[0071] [Fig.7a] illustrates an example of spatial representation 80a of a strand skeleton obtained using the above method using 5 question points P. The dashed lines I represent standard interpolations (quadratic, cubic and radial basis function) of the 5 question points P. It appears in [Fig.7a] that the skeleton 90a found using the method 100 according to the first aspect of the invention makes it possible to find a good reconstruction of the skeleton with only 5 question points, which is not possible using the standard interpolation techniques giving the curves I.

[0072] [Fig.7b] illustrates an example of spatial representation 90b of a strand skeleton obtained using the method according to the first aspect of the invention using 8 question points P. This figure shows that standard interpolation methods require at least 8 points to correctly describe the path of the strand.

[0073] According to another embodiment of the method 100 according to the first aspect of the invention, the representation learning technique used during step 110 is an encoder-type machine learning model.

[0074] According to this embodiment, the latent variables have a form of type y _ / ' ùA X being a latent representation of a skeleton belonging to the J encoder \ J database. The function f is therefore used to reconstruct a spatial representation of the strand from manually identified points. The learning phase therefore consists of identifying the hyper parameters of the model, for example the weights of the neural network representing the autoencoder, making it possible to minimize the difference between X and [A ( X) V The autoencoder is therefore an alternative to principal component analysis as a representation model for approximating the skeleton points.

[0075] According to another embodiment, the representation learning technique uses a recurrent neural network or RRN according to the English terminology "Recurrent Neural Network". In this case the set of points belonging to the skeleton of the strand is a time series of xt coordinates.

[0076] The spatial representation of the skeleton of the strand has the form 7 ~ y with a series of entry points xi, a series of exit points zt, and A the corresponding latent space. Thus, the t RNN function is applied to the current input but provides a solution according to the cumulative conditions of the previous inputs. The fRNN function is a recurrent learning model, for example a recurrent neural network. An alternative to the recurrent neural network is any prediction approach (or "forecasting" in English terminology) which, from a sequence of strand centers (x0,..xn) allows to make the prediction (xn+i,..x n+m)*

[0077] In the case of three-dimensional reconstruction of the skeleton of the strand, it is possible to consider two one-dimensional models y æ fw and f L; ,,, \ or a single two-dimensional model 7 ~ v Y

[0078] [Fig.8] illustrates an embodiment of the device 90 for determining a spatial structure of a skeleton of a strand according to an embodiment of the invention. The device 90 comprises an input interface 91 and a calculation circuit such as a microprocessor 92.

[0079] The input interface is configured to receive a set of latent variables VL and a set of spatial coordinates z(x) of the strand whose structure is to be determined.

[0080] The circuit or microprocessor 92 is configured to determine the spatial representation of the skeleton of the strand from the set of spatial coordinates 4-x), the spatial representation being a function of the latent variables.

[0081] According to one embodiment, the device 90 is configured to implement the method 100 according to the first aspect of the invention.

[0082] Figure 9 represents a device 901 for determining a latent space of a database D comprising spatial representations of strand skeletons extending in a predetermined direction x.

[0083] The device 901 comprises an input interface 910 for receiving the database D.

[0084] The device 901 further comprises a circuit or microprocessor 920 for obtaining the latent space by applying a representation learning technique to the database and an output interface 930 for providing a set of latent variables VL representing the database in the latent space.

Claims

Claims

1. A computer-implemented method (100) for determining a spatial representation of a skeleton of a strand of a portion of a piece made of woven composite material, the strand extending in a predetermined direction1, said method comprising: • Obtaining (110), from a database comprising spatial representations of skeletons of strands extending in the predetermined direction x, a latent space representative of the database, the database being described in the latent space by a set of latent variables; • Receiving (120) spatial coordinates = of Nq points of the skeleton of the strand located at respective positions along the predetermined direction x; • Determining (130) the spatial representation of the skeleton of the strand from the obtained spatial coordinates Z(i, said spatial representation of the skeleton of the strand being a function of the latent variables.

2. Method according to the preceding claim in which the latent space of the database is obtained using a representation learning technique chosen from a list comprising: principal component analysis, auto-encoder, recurrent neural networks.

3. Method according to one of the preceding claims in which the part made of woven composite material is an aeronautical part.

4. Method according to one of the preceding claims in which: • Obtaining the latent space representative of the database comprises a principal component analysis of said database, the latent variables being P principal components mi; • The spatial representation of the skeleton of the strand comprises a set of points associated with respective coordinates g(.x), in which 2(x) is equal to a linear combination of type ■ w(x)' with a' real coefficients.

5. Method according to the preceding claim in which the coefficients ai are determined by minimizing a deviation between the received spatial coordinates Zci and the values z(Xq) of the linear combination evaluated for the positions

6. Method according to the preceding claim in which the minimized deviation is of the form < ri 2.

7. Method according to one of claims 1 to 3 in which: • Obtaining the latent space representative of the database comprises the application of an autoencoder type model to the database; • The latent variables of the database are of type y — f / SA X being a spatial representation of an encoder skeleton belonging to the database.

8. Method according to the preceding claim in which obtaining the latent variables Y from the database comprises determining two functions / .and / , , chosen so as to minimize the gap encode aecoder between X and / ■ / AA\j. J decode \V encode \ / /

9. Method according to one of claims 1 to 3 in which obtaining the latent space representative of the database comprises the application of a recurrent neural network type model.

10. Device for determining a spatial representation (S) of a skeleton of a strand of a portion of a piece made of woven composite material, the strand extending in a predetermined x direction, said device comprising: • An input interface (91) for receiving: • Latent variables (VL) describing a latent space of a database comprising spatial representations of strand skeletons extending in the predetermined x direction; • A set of spatial coordinates z(x) of the strand; • A circuit (92) configured to determine the spatial representation (S) of the skeleton of the strand from the set of spatial coordinates <(A the spatial representation being a function of the latent variables.

11. Device (901) for determining a latent space of a database comprising spatial representations of strand skeletons extending in a predetermined x direction, said device comprising: • An input interface (910) for receiving the database (D); • A circuit (920) for obtaining the latent space by applying a representation learning technique to the database; • An output interface (930) for providing a set of latent variables (LV) representing the database (B) in the latent space.

12. Computer program product comprising the instructions for implementing the method according to one of claims 1 to 9 when this program is executed by a processor.