METHOD AND DEVICE FOR PROCESSING A THREE-DIMENSIONAL IMAGE FOR INSPECTING A PART AND METHOD FOR INSPECTING A PART USING THE PROCESSING METHOD

Abstract

One aspect of the invention relates to a method for processing a three-dimensional image associated with a plurality of transverse images and comprising a part to be checked, comprising for each associated transverse image the following steps: Binarization (E311) of the transverse image concerned to obtain a mask formed of a plurality of pixels corresponding to the pixels of the transverse image relating to the section of the part, Determination (E3121, E3122), from the mask, of a surface representing said section, Determination (E3123) of a parametric function describing said surface, Definition of a constant displacement field on each column of pixels and dependent on the column coordinates of the pixels so as to apply a flattening transformation to the surface represented by the parametric function,Straightening (E314) of the section of the part in the cross-sectional image concerned by applying the determined displacement field to the pixels of said image. Figure to be published with the abstract: Figure 7,

Description

Title of the invention: METHOD AND DEVICE FOR PROCESSING A THREE-DIMENSIONAL IMAGE FOR INSPECTING A PART AND METHOD FOR INSPECTING A PART USING THE PROCESSING METHOD TECHNICAL FIELD OF THE INVENTION

[0001] The technical field of the invention is that of non-destructive testing of industrial parts using three-dimensional tomographic images, and more particularly that of non-destructive testing of parts made from a woven composite material.

[0002] The present invention relates to a method for processing a three-dimensional tomographic image within which a part is defined, and in particular to a processing method comprising rectification. The invention also relates to an associated control method.

[0003] The invention also relates to an associated device for processing a three-dimensional tomographic image within which a part is defined. Finally, it relates to an associated computer program.

[0004] The present invention finds an advantageous application for the visual inspection of an aeronautical part made of three-dimensional woven composite material. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0005] Aeronautical parts made of three-dimensional woven composite material, for example engine fan blades, are critical parts that must be systematically checked in their entirety to ensure their conformity and / or material health. In particular, the weave must be systematically analyzed in the volume of the composite material, in order to ensure that it does not contain defects or anomalies.

[0006] Weaving refers to the woven textile framework of the three-dimensional woven composite material, called reinforcement. The reinforcement is composed of threads, or strands, woven according to a defined weaving topology over several weaving layers. In each weaving layer, the strands are woven in two main orthogonal directions, called warp direction and weft direction. The weaving layers are superimposed in a third direction corresponding to the thickness of the textile framework and called reinforcement direction. These three directions of warp, weft and reinforcement preferably form an orthogonal reference frame, called the part reference frame.

[0007] Currently, the analysis of the weaving of the part is carried out in a non-destructive manner from a tomographic volume obtained by X-ray tomography imaging.

[0008] The tomographic volume is a three-dimensional image representing the part.

[0009] The analysis of the weaving consists first of extracting from this three-dimensional image stacks (or sequences) of two-dimensional cuts (or sections). In each stack, each two-dimensional cut corresponds to a section of the three-dimensional image along a section plane perpendicular to the same axis of the part's reference frame. The section planes are distinct and successive along this axis of the part. This axis is different for each stack of two-dimensional cuts. For example, the two-dimensional cuts of the first stack are perpendicular to the thickness direction while the two-dimensional cuts of the second and third stack are respectively perpendicular to the warp direction and the weft direction.

[0010] Visual analysis then consists of visually analyzing the stacks of two-dimensional sections. Visual analysis of a given stack consists of manually and successively scrolling through each two-dimensional section of this stack to follow a given strand or a given set of strands in its entirety. This thus makes it possible to detect a difference in the position of the strands, which would be the sign of a potential anomaly.

[0011] Some aeronautical parts have a very aerodynamic shape. Thus, fan blades include marked curvatures. The visual analysis then consists of scrolling through the two-dimensional sections of certain stacks several times. For example, the two-dimensional sections of the first stack (which, according to the example given previously, each represent a weaving layer) are displayed in the successive order of the sections and in the reverse order. This multiple scrolling makes it possible to compensate for the fact that certain two-dimensional sections do not give a view of the entire weaving layer due to the curvatures of the part.

[0012] To avoid having to repeatedly scroll through the two-dimensional sections, it is known from document FR3047339A1 to carry out pre-processing of the three-dimensional image aimed at erasing, or flattening, the curvatures of the part.

[0013] This pre-processing is carried out by straightening the part represented in the three-dimensional image corresponding to the tomographic volume. More precisely, a straightened three-dimensional image is reconstructed all around a median surface of the part. The straightening model then consists of unrolling the median surface of the part in order to lay it flat. This straightening model has the effect of preserving the length of the median surface of the part.

[0014] Such rectification is anamorphic: during reconstruction, the regions of the part are not interpolated in the same way depending on their distance from the surface. For example, the regions furthest from the surface, typically the regions located outside the part (or "extrados") and the regions located inside the part ("intrados") are undersampled or oversampled, respectively. As a result, they appear respectively spread or squashed on the straightened three-dimensional image.

[0015] More generally, due to this unbalanced interpolation, regions of the part in the rectified three-dimensional image may appear blurred or exhibit artifacts. These regions cannot then be effectively controlled. Summary of the invention

[0016] The present invention provides an alternative pre-processing method for improving the quality of the rectified three-dimensional image.

[0017] More particularly, a first aspect of the invention relates to a method for processing a three-dimensional image for the inspection of a part included in the three-dimensional image, the three-dimensional image being associated with a plurality of two-dimensional transverse images, each transverse image corresponding to a section of the three-dimensional image along a section plane perpendicular to a main axis of said part, the section planes being distinct, each transverse image comprising a section of the part, each transverse image being formed of a plurality of pixels, the method comprising a rectification processing comprising the following steps: • Processing each transverse image to obtain a plurality of straightened transverse images, each straightened transverse image comprising a straightened section of the part, said processing comprising, for each transverse image, the following steps: • Generation, from the transverse image concerned, of a binary image to obtain a mask formed of a plurality of pixels corresponding to the pixels of the transverse image relating to the section of the part, the plurality of pixels of the mask being arranged according to lines of pixels and columns of pixels, each pixel being defined by a line coordinate and a column coordinate, • Determination, from the mask, of a surface representing the section of the part, • Determination of a parametric function describing said surface, • Definition of a constant displacement field on each column of pixels and dependent on the column coordinates of the pixels so as to apply a flattening transformation to the surface represented by the parametric function, • Straightening the section of the part in the relevant cross-sectional image by applying the determined displacement field to the pixels of the relevant cross-sectional image, to generate the straightened cross-sectional image associated with the relevant cross-sectional image, • Reconstruction of a straightened three-dimensional image containing the straightened part from the straightened cross-sectional images obtained from the raw cross-sectional images.

[0018] Thanks to the displacement field which defines relative displacement vectors, parallel to each other, the flattening model is based on a simple transformation, consisting of sliding portions of a curved section relative to each other. The portions of the straightened section are thus simply moved (translated).

[0019] This displacement model finds an advantage during the step of reconstructing the three-dimensional image from the rectified transverse images. Indeed, the displacement model is isochoric, that is to say that the volume of the part is preserved during the rectification processing process. This conservation of the volume makes it possible to limit the interpolation errors of the rectified three-dimensional image. The rectified three-dimensional image thus advantageously presents a homogeneous quality over the entire rectified part. This allows a reliable and complete inspection of the part.

[0020] In addition to the characteristics which have just been mentioned in the preceding paragraph, the method according to the first aspect of the invention may have one or more complementary characteristics among the following, considered individually or according to all technically possible combinations: • The surface area representing the section of the part is determined by carrying out the following steps: • Defining a predetermined image having the same dimension as the binary image, the predetermined image comprising a plurality of pixels, each pixel having a predetermined value corresponding to the row coordinate of said associated pixel in the predetermined image, • Assigning a predefined value to each pixel of the mask, said predefined value corresponding to the value of the pixel in the predetermined image having the same row and column coordinates as the pixel of the mask concerned, and • For each column of pixels in the mask, determining a center of said column from the predefined pixel values and the number of pixels in said column, and assigning the center determined at each pixel of said column, the centers corresponding to a representation of said surface. The centers are determined from the result of a matrix product between a matrix representing the predetermined image and a matrix representing the binary image. The centers are barycenters, each barycenter being determined on the basis of the following expression: ^yJt(x,y)*B(x,y) where: HyB^y) • x- y are respectively the column and row coordinates of each pixel of the mask, • IP(x,y) is the matrix representing the predetermined image, • B(x, y ) is the matrix representing the binary image. The parametric function is composed of a spline. The application of the displacement field is broken down into a translation operation and / or an interpolation operation. The processing method comprises a step of determining a projection reference frame, the steps of defining the parametric function and the displacement field being a function of the determined projection reference frame so that: • after the step of determining the surface and before the step of determining the parametric function, the surface is determined in the projection frame using a system of frame change equations defined on the basis of an angle parameter, • after the step of determining the displacement field and before the rectification step, the pixels of the transverse image concerned are interpolated in the projection reference frame to obtain a projected transverse image, the rectification step then being implemented on the projected transverse image. The angle parameter has a predetermined non-zero value. The angle parameter is determined between the binary image generation step and the surface determination step in the projection frame, by performing the following steps: • Extraction of pixels relating to the centers or relating to the section of the part in the binary image, said pixels forming a point cloud, • Identification of a preferred direction of the point cloud in the binary image reference frame, • Determination of the angle parameter from the identified preferred direction. • The preferred direction is identified using a principal component analysis of the point cloud along two principal axes, the axes being perpendicular, the preferred direction being identified by selecting from the eigenvectors produced by the principal component analysis the eigenvector associated with the largest eigenvalue, the angle parameter then being determined from the components of the selected eigenvector and limiting the value range of said parameter to the range [-45°, +45°]. • The step of determining the angle parameter is implemented in parallel with the step of determining the surface produced from the mask. • The plurality of rectified transverse images is obtained simultaneously, by parallel processing of each transverse image of the plurality of transverse images. • The step of generating the binary image comprises a step of selecting, in the binary image, a subset of pixels representing a region of interest comprising the section of the part, the mask being obtained from the subset of selected pixels. • The step of reconstructing a rectified three-dimensional image comprises a step of smoothing, along the main axis of the part, the rectified three-dimensional image, the smoothing step comprising the following steps: • Determination, from the rectified three-dimensional image, of a smoothing function of the rectified section of each rectified transverse image forming the rectified three-dimensional image, • Determination of a weighting function of the displacement field associated with each rectified transverse image, using the smoothing function and the displacement field used to obtain each rectified transverse image, • Application of the displacement field weighting function on each straightened transverse image to obtain a smoothed three-dimensional image.

[0021] A second aspect of the invention relates to a method for controlling a part from a three-dimensional image, comprising the steps of the processing method according to the method which is the subject of the first aspect of the invention to obtain a rectified three-dimensional image, and a step of controlling the part from the rectified three-dimensional image.

[0022] The part may be made of a three-dimensional woven composite material.

[0023] A third aspect of the invention relates to a computer program comprising instructions which, when the program is executed by a computer, lead the latter to implement the steps of the treatment method according to the first aspect of the invention.

[0024] A fourth aspect of the invention relates to a device for processing a three-dimensional image for the inspection of a part included in the three-dimensional image, the three-dimensional image being associated with a plurality of two-dimensional transverse images, each transverse image corresponding to a section of the three-dimensional image along a section plane perpendicular to a main axis of said part, the section planes being distinct, each transverse image comprising a section of the part, each transverse image being formed of a plurality of pixels, the device comprising a processor configured to implement a rectification processing comprising: • Processing each transverse image to obtain a plurality of straightened transverse images, each straightened transverse image comprising a straightened section of the part, said processing comprising, for each transverse image, the following steps: • Generation, from the transverse image concerned, of a binary image to obtain a mask formed of a plurality of pixels corresponding to the pixels of the transverse image relating to the section of the part, the plurality of pixels of the mask being arranged according to lines of pixels and columns of pixels, each pixel being defined by a line coordinate and a column coordinate, • Determination, from the mask, of a surface representing the section of the part, • Determination of a parametric function describing said surface, • Definition of a constant displacement field on each column of pixels and dependent on the column coordinates of the pixels so as to apply a flattening transformation to the surface represented by the parametric function, • Straightening the section of the part in the relevant cross-sectional image by applying the determined displacement field to the pixels of the relevant cross-sectional image, to generate the straightened cross-sectional image associated with the relevant cross-sectional image, and • Reconstruction of a straightened three-dimensional image containing the straightened part from the straightened transverse images.

[0025] 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

[0026] The figures are presented for information purposes only and in no way limit the invention. • [Fig.l] represents, in schematic form, a perspective view of an aeronautical part, • [Fig.2] represents, in the form of a flowchart, an example of a process processing a three-dimensional image in accordance with the invention, • [Fig.3A] represents, in schematic form, a first view of an example of a three-dimensional image of the part used by the treatment method, • [Fig.3B] represents, in schematic form, a second view of an example of a three-dimensional image of the part used by the treatment method, • [Fig.3C] represents, in schematic form, a third view of an example of a three-dimensional image of the part used by the treatment method, • [Fig.4] represents, in schematic form, an example of an image raw cross-section obtained from the raw three-dimensional image at the end of the first step of the process shown in [Fig.2], • [Fig.5] represents, in the form of a flowchart, the main stages of the second step of the process shown in [Fig.2], • [Fig.6] represents, in the form of a flowchart, a first sub-step of the second step of the process shown in [Fig.2], • [Fig.7] represents, in schematic form, the first sub-stage of the second step of the process shown in [Fig.2], • [Fig.8] represents, in the form of a flowchart, a variant of implementation of the step represented in [Fig.5], • [Fig.9A] represents an example of a raw longitudinal image, • [Fig.9B] represents the rectified longitudinal image corresponding to the raw longitudinal image represented in [Fig.9A], • [Fig. 10A] represents a first example of a longitudinal image associated with a three-dimensional image rectified by the method represented in [Fig.2], • [Fig.10B] represents a second example of a longitudinal image associated with a three-dimensional image rectified by the method represented in [Fig.2], • [Fig.10] represents a third example of a longitudinal image associated with a three-dimensional image rectified by the method represented in [Fig.2], • [Fig.10D] represents a fourth example of a longitudinal image associated with a three-dimensional image rectified by the method represented in [Fig.2], • [Fig. 11] represents, in the form of a flowchart, a variant of implementation of the process, • [Fig. 12] represents, in the form of a flowchart, an alternative mode of implementation of the process shown in [Fig.2], • [Fig. 13] represents, in schematic form, a step of the alternative implementation mode of the process shown in [Fig. 12], • [Fig.l4A] represents, in the form of a flowchart, a first variant of the alternative implementation mode represented in [Fig. 12], • [Fig.l4B] represents, in the form of a flowchart, a second variant of the alternative implementation mode represented in [Fig. 12], • [Fig. 15] represents, in the form of a flowchart, a variant of implementation of the method shown in [Fig. 12], • [Fig. 16] represents, in schematic form, a device making it possible to implement the method of processing a three-dimensional image in accordance with the invention. DETAILED DESCRIPTION

[0027] The present invention is placed in the context of the non-destructive testing of an aeronautical part made from a woven composite material, using three-dimensional tomographic images. More particularly, the invention aims to allow rapid and easy visual inspection of the weaving topology of this part. In particular again, the invention aims to provide three-dimensional images, pre-processed by rectification, of better quality than the state of the art, thus making it possible to improve the reliability of the inspection.

[0028] The invention finds in particular an advantageous application for the control of fan blades. For example, the invention finds an advantageous application for the control of fan blades of LEAP engines for "Leading Edge Aviation Propulsion" according to the commonly used acronym of Anglo-Saxon origin.

[0029] In the remainder of the description, and for the sake of simplicity, the term “part” will be used to designate both the physical part 10 and the representation (or model) of the part.

[0030] The material aeronautical part is intended for use in the field of aeronautics, more precisely in the context of turbomachines. The following description, as well as the figures, are illustrated for an engine fan blade. Naturally, other aeronautical parts may be concerned by the invention.

[0031] [Fig.l] represents a perspective view of an aeronautical part 10 such as a fan blade 10.

[0032] The part 10 is represented in an orthogonal XYZ reference frame defined by computer-aided design software, or CAD. In this reference frame, the Z axis corresponds to the largest dimension of the part 10. This Z axis is associated with a longitudinal direction of the part 10 (it is called “longitudinal Z axis” in the following). The XY planes orthogonal to the longitudinal Z axis are transverse planes.

[0033] In the present description, the term “length” is used to designate the dimension of the part 10 along the longitudinal axis. The term “thickness” is used to designate the dimension of the part 10 along the Y axis. In addition, the term “section” designates the shape of the part included in a plane perpendicular to one of the axes X, Y, Z of the part 10 and intersecting the volume of this part 10.

[0034] The part 10 is made of three-dimensional (or 3D) woven composite material.

[0035] A 3D woven composite material is an assembly comprising at least one woven textile framework called reinforcement and a binder called matrix. The manufacture of a part in a woven composite material requires a first step of making the reinforcement by weaving, then a second step of assembly with the matrix, for example by injection after shaping in a mold.

[0036] The reinforcement is composed of threads, or strands, woven in two main orthogonal directions, called warp direction and weft direction, and in a third direction corresponding to the thickness of the weaving. In the XYZ coordinate system of the part 10, the warp direction is aligned along the longitudinal axis Z, the weft direction is aligned along the X axis, and the third direction is aligned along the third axis Y.

[0037] In the remainder of the description, the term "warp strands" will refer to the strands aligned along the longitudinal axis Z of the XYZ reference frame of the part 10, and the term "weft strands" will refer to the strands aligned along the X axis.

[0038] With reference to [Fig.l], the part 10 has an aerodynamic shape. Thus, the part 10 has a high aspect ratio, that is to say a large dimension along the longitudinal axis Z compared to the transverse dimensions. It also has curvatures which extend along the longitudinal axis Z, on all intrados Int and extrados Ext surfaces. In particular, it has a crescent-shaped lunula-shaped curvature linked to the intrados-extrados Int, Ext profiles and a helical shape. Finally, its thickness in the X direction varies greatly along the longitudinal axis Z.

[0039] [Fig. 2] represents, in the form of a flowchart, an example of a processing method 1 in accordance with the invention. This processing method 1 is also referred to as “method 1” in the following.

[0040] The steps of this processing method 1 are typically carried out using a calculation unit 100, illustrated in [Fig. 16], comprising data processing means 110 (such as a processor) and a memory 120. A display monitor 130 may be provided, configured to display data in particular from the calculation unit. More generally, the processing method 1 according to the invention is here implemented by computer.

[0041] Prior to implementing the method 1, a method for acquiring a three-dimensional image I3D of the part 10 is implemented. This method for acquiring the three-dimensional image I3D of the part 10 is for example implemented in practice using an X-ray tomographic imaging system and computing means.

[0042] The reference frame of the three-dimensional image I3D corresponds to the global reference frame XYZ of the part 10.

[0043] The three-dimensional image I3D is composed of voxels with coordinates (x,y,z), i.e. three-dimensional pixels. Each voxel is associated with a gray level value corresponding to a local X-ray attenuation coefficient.

[0044] In practice, the tomographic representation of the part 10 is defined, within the three-dimensional image I3D, on the basis of voxels having particular gray levels. These gray levels are a function, for example, of the local density of the material forming the part and of the atomic number of the elements which constitute this material.

[0045] In the present example, in order to obtain the desired spatial resolution, the three-dimensional image I3D is obtained from three local three-dimensional images representing three consecutive regions of the part. The resulting three-dimensional image I3D is then called a global three-dimensional image. The number of regions of the part is not limited to three.

[0046] [Fig.3A], [Fig.3B] and [Fig.3C] represent different views of a three-dimensional image I3D of the part 10.

[0047] In these figures 3A, 3B and 3C, the three local three-dimensional images are referenced I3Dji, I3D>2 and I3D>3. A local reference frame is associated with them. This local reference frame is distinct from the global XYZ reference frame of the part 10. This local reference frame is preferably defined on the basis of the part 10, so that the weaving directions (warp, weft and thickness) are best aligned with the axes of the local reference frame.

[0048] The processing method 1 is applied to the global three-dimensional image I3D or to each local three-dimensional image I3Dji, I3D>2 and I3D>3.

[0049] In the remainder of the description, the term “raw” three-dimensional image J^D refers to the overall three-dimensional image I3D or one of the intermediate three-dimensional images I3D>b I3D>2 and I3D>3. The term “raw” refers to the three-dimensional image that has not yet undergone processing.

[0050] The processing method 1 is now described in detail.

[0051] As shown in Figure 2, the method 1 comprises a first step E2 aimed at providing a set of several raw transverse two-dimensional images y*^ i — 1 ■ n to which a rectification processing E3 will be applied.

[0052] These raw transverse images y^. are associated with the raw three-dimensional image jb. They each correspond to a different transverse XY section Si of the part 10, each XY section corresponding to a coordinate on the longitudinal axis Z distinct from that of the other XY sections. As described previously, the term “transverse” here refers to a transverse XY section plane orthogonal to the longitudinal axis Z.

[0053] As shown in Figure 4, each raw transverse image y^y corresponds to an image formed from a plurality of pixels Pj(x, y) distributed in rows and columns. For the sake of readability, the raw transverse image here represented in [Fig.4] is not a real image but a schematic representation of this real image.

[0054] Conventionally, the raw transverse image y* is defined on the basis of a “grid” formed of pixels p}( x, y ). Each pixel P^x, y) has a column coordinate and a row coordinate x. These row and column coordinates are defined relative to an origin O with coordinates (0,0). In Figure 4, the origin O is here defined at the top left corner of the raw transverse image Jb.vv. lyAÏ

[0055] Here, as visible in Figure 4, each pixel Pj(x, y ) is associated with a gray level. The section Si of the part 10 is defined in the raw transverse image y^y by pixels having particular gray levels. These gray levels are in fact different from the gray levels of the pixels not relating to the part 10. In addition, they extend over a greater dynamic range of gray levels.

[0056] In Figure 4, it is also visible that the section Si has a crescent shape oriented along the pixel lines and delimited by a lower curved line C](x) and an upper curved line. The terms “lower” and “upper » are here chosen with respect to the reference (X y) illustrated in figure 4, and defined for the raw transverse image lyÀ. 1 '

[0057] An objective of the straightening treatment E3 which will be described below is to straighten the section Si of the part 10 by first determining a mean curved line between the lower and upper curved lines c and Pu's by making this mean curved line substantially rectilinear.

[0058] As will also be described later, this E3 straightening treatment will be applied identically to each raw transverse image. All of the straightened sections will then compose a straightened part on which the weaving analysis can be carried out.

[0059] The raw transverse images jb are obtained in the following manner.

[0060] In a first sub-step E21, a set of different coordinates Z-, ï — 1 : II, along the longitudinal axis Z is determined in the raw three-dimensional image T1*. These coordinates are preferably chosen to increase along the longitudinal axis Z in a range of values extending over the length of the part 10 represented in the raw three-dimensional image. Thus, the coordinates z; are distinct from one another.

[0061] Then, for each of these coordinates z? i = 1 : 11, the corresponding section is determined along a cutting plane XY orthogonal to the longitudinal axis Z. As the coordinates z; are distinct from each other, the sections are also distinct.

[0062] Finally, in a step E22, the transverse images i — 1; “corresponding to the determined sections are generated.

[0063] In the remainder of the description, the transverse images obtained at the end of step E2 are called “raw transverse images” because they have not yet undergone the rectification processing of step E3.

[0064] At the end of step E2, we thus have a plurality of raw transverse images I!'!YY i = [ • r, to which the rectification processing described below will be applied. In the following, the rectification processing is described for a transverse image jb but applies in the same way for all the raw transverse images i — 1 ' II-

[0065] The rectification processing of step E3 aims to obtain a rectified three-dimensional image from a plurality of rectified sections S; of the part 10.

[0066] [Fig.5] represents, in the form of a flowchart, the main steps of step E3.

[0067] The first step E31 is described below in relation to Figures 6 and 7. As indicated previously, this first step E31 is carried out on each transverse image fi^r obtained in step E2. It specifically aims to implement the straightening of the section Si of the part 10 represented in the raw transverse image „ concerned.

[0068] To do this, we first determine a curve f^x ) which makes it possible to represent the average curved line of the section Si of the part 10, then we determine a displacement field « / (x) which makes it possible to flatten this curve x Finally, we carry out the straightening of the section Si of the part 10 with the displacement field ufix) determined.

[0069] These different operations are represented respectively by steps E311, E312 and E313 described below in relation to [Fig.6].

[0070] The determination of the curve fÀ x ) representing the average curved line of the section of the part comprises the successive steps E311 and E312.

[0071] First of all, in the first step E311 illustrated in FIG. 6, a binarization (i.e. a conversion into binary form) of the raw transverse image fi concerned is carried out to identify all the pixels relating to the section of the part 10. A binary image B is then obtained, formed of a plurality of pixels and having the same dimensions as the raw transverse image concerned. In this binary image B^ all the pixels relating to the section Si of the part 10 represented in this raw transverse image have the value 1, and all the other pixels have the value 0. Thus, the binary image Bt provides a mask Af / jyy relating to the section of the part in the raw transverse image fi.^ concerned.

[0072] The mask M^y obtained is an image as illustrated in figure 7 (in connection with step E311). For the sake of readability, this image is represented here not in the form of a real image but of a schematic representation of this real image. Similarly, the other images of figure 7 (in connection with steps E312 and E314) are schematic representations of real images. It is formed of pixels and has the same dimensions as the raw transverse image. Furthermore, each pixel of the mask M^y is then associated, by definition, with a value 0 or 1, in the manner described in the preceding paragraph. On the image of the mask M^y represented in [Fig.7], the pixels having a value 0 (not relating to the section of the part) are represented in black, while the pixels having a value 1 (relating to the section of the part) are represented in white. For readability, the black color representing pixels not related to the section of the part will be replaced by white on the other images in [Fig.7].

[0073] Binarization can be performed using gray levels, by choosing a threshold. The choice of the threshold can advantageously be obtained via the Otsu method. More details concerning this method are for example given in the document “A threshold selection method from gray-level histograms”, IEEE Trans. Sys. Man. Cyber., vol. 9, 1979, pp. 62-66, by Nobuyuki Otsu.

[0074] Advantageously, the mask M^y obtained can be smoothed. This smoothing can be carried out by performing morphological operations of the erosion and / or dilation type on the mask M^y.

[0075] Once the mask M^y is obtained from step E311, the method 1 comprises a step E312 of determining the curve f^x). This step is carried out from the mask ^iAY-

[0076] In practice, this curve f^x) corresponds to an average curve f Mi(x), that is to say a curve which is equidistant from the lower and upper curved lines ^(x), c2(x) (in the Y direction of the columns of pixels of the mask M^y, cf. [Fig.4]).

[0077] The average curve f ^.{x, y) is preferably determined in the following manner.

[0078] First of all, in step E3121 illustrated in figure 7, an image called “predetermined image Pi” is defined. This has the same dimension as the binary image Bj. Each pixel Pp{x, y) which makes up the predetermined image Pi has a value which corresponds to the line coordinate T of this pixel. In other words, the predetermined image Pi is an image of the coordinates of the lines of the mask: Pi(x, y} - y. On the predetermined image P; illustrated in [Fig.7], these values of the coordinates of the lines of the mask are represented schematically by different densities of points.

[0079] Then, in step E3122 also illustrated in FIG. 7, each pixel PM(X y) of the mask is assigned the value of the corresponding pixel pp(x, y ) of the predetermined image P^ By definition, a pixel PM(X, y) of the mask which corresponds to a pixel Pp( x,y) of the predetermined image has in the mask the same row coordinate y and the same column coordinate x as this pixel pp{x, y ) of the predetermined image in the predetermined image.

[0080] The image G{ obtained at the end of this step E3122 differs from the mask only in that the pixels which had the value 1 in the mask now have, as their value, the value of their line coordinate y. The image G{ obtained is in other words a image of the line coordinates Gj(x, y) - y. This image is illustrated in [Fig.7] in relation to step E3122.

[0081] Note that each pixel of the image G( of the line coordinates relating to the section Si of the part 10 has been weighted by its line coordinate y This weighting is now used to determine a center ym of each column of pixels of the mask

[0082] In practice, we determine a barycenter ym of each column of pixels of the mask

[0083] To do this, the following processing is performed on each column of the image G{.

[0084] We first determine a ratio between the sum of the values of the pixels of this column and the number of pixels in the column belonging to section Si of part 10. This ratio gives the barycenter ym of the column concerned.

[0085] Then, the determined barycenter is assigned to the pixels of the relevant column of the mask M,. The mask M, has thus been modified: the pixels which previously had the value 1 now have the value of the barycenter of their column ym as their value. The modified mask is, in other words, an image C^ X, y ) = y of the column centers of the mask

[0086] Preferably, each column center ym is determined from a matrix product between a matrix Bj(x, y) representing the binary image Bj and a matrix Pi(x, y) representing the predetermined image P,.

[0087] More precisely, each column center ym is the result of the following ratio:

[0088] • Where x' y are respectively the column and row coordinates of each pixel of the mask M^y, • P^x, y) is the matrix representing the predetermined image P, ( x, y ), and • B{ ( x, y ) is the matrix representing the binary image B{ (x, y).

[0089] We further note that (x, y)*Bi(x, y) is the sum of the values of the pixels in column E and that ^yB^x, y) is the number of pixels in the column belonging to section Si of part 10.

[0090] Finally, in step E3123, a parametric function f^x) is determined which describes the centers ym of the columns of the image Cj of the centers. The parametric function f^x) thus determined forms the average curve fMi(x) sought.

[0091] This average curve fMi(x) is illustrated in figure 6, in relation to step E3123. It is represented in an image of the same dimension as the binary image B;.

[0092] The parametric function fx ) is defined to be constant over each column -v of pixels of the image of the centers and to depend on the column coordinates x.

[0093] To do this, we implement an interpolation on the image Q of the column centers ym to find the parameters of the parametric function fÂx).

[0094] These parameters are then adjusted progressively on pairs {column coordinate, column center ym associated with this column coordinate] so that the parametric function tends towards the column centers: f ^x) ~ ym. This adjustment is for example implemented by minimizing an objective function such as the L2 norm of the error between the value of the center and a value of the parametric function - Il y - / (^)11 • The optimization ends when a convergence criterion is reached, or when a certain number of iterations is reached.

[0095] Preferably, the parametric function f^x) is composed of a spline.

[0096] The spline is constructed to correspond to the best approximation of the pairs of points {column coordinate, column center y„t associated with this column coordinate] obtained from the centers (or barycenters). Each pair of points corresponds to a center (or a barycenter).

[0097] It is possible to use other parametric functions, for example polynomial functions. It should be noted, however, that the use of a parametric function composed of a spline offers the advantage of avoiding unwanted oscillations at the ends (i.e. at the edges, along the upper and lower curved lines c^x), c2(x) of the mask jij.y

[0098] Advantageously, the use of a spline is associated with a smoothing effect which makes it easier to converge the adjustment of the parameters of the function. In addition, a smoother and more regular function f^x) is obtained which facilitates the following step E313 of determining a displacement field.

[0099] Alternatively, it is possible to determine a median curve instead of a mean curve to represent the surface of the section of the part. In this case, median centers will be determined instead of barycenters.

[0100] It should be noted that this way of determining the average curve f does not rely on a detection of the external contours of the mask M^y- This contour detection generally comprises a step of detecting the upper and lower contours and a step of determining the average resulting from these two extracted contours. This contour detection has the disadvantage of being particularly sensitive to the presence of imperfections in the binary image, for example to the presence of noise on the contours (aberrant points) or discontinuities (missing points). However, these imperfections are common in practice. Thus, the presence of external parts to the part to be controlled, for example support parts to hold the part, can interfere and make the contours discontinuous.

[0101] Advantageously, the manner of determining the average curve in step E312 makes it possible to overcome these drawbacks and, thus, to be much more robust than contour detection. Indeed, by calculating the column centers ym from the predetermined image P^, many more points are available than when using only the two points belonging to the internal and external contours. The determination of the centers is thus possible even if one of the contours has a missing point. It is also less sensitive to the presence of an aberrant point.

[0102] Furthermore, determining the centers by matrix calculation is much more advantageous from a calculation time point of view since computer languages and computers are optimized for this type of operation.

[0103] The method continues with a step E313 of determining the displacement field u^x) carried out from the average curve f( x ) determined previously. The displacement field u^x) thus determined is illustrated in [Fig.7] in relation to step E314.

[0104] The displacement field u^x) gives the relationship between the mean curve fx ) and a flattened mean curve Such a flattened mean curve f MiXx) x» 4 rll j VJL fü is for example illustrated in figure 7 in connection with step E314. The flattened average curve f is for example obtained by projection of the average curve f Mi(x) onto the abscissa axis x.

[0105] To determine this displacement field, the inventors have identified that an important parameter to take into account is the way in which the part 10 deforms. Observations carried out within the framework of the invention have thus made it possible to show that the part 10 deforms as a structure where essentially the weft strands (directed in the direction of the pixel lines in the mask) move in simple shear parallel to the thickness Y of the part 10 (or perpendicular to the average warp Z and weft X directions).

[0106] Thus, the inventors sought a displacement field u^x) which makes it possible to reproduce these observed deformations. In particular, the inventors sought a displacement field which makes it possible to preserve the volume of the part 10 during the straightening treatment. In other words, the inventors sought a displacement field such that the straightening treatment is an isochoric transformation.

[0107] Such a displacement field u^x) defines relative displacement vectors, parallel to each other and directed according to the direction of the pixel columns of the mask (which corresponds to the direction orthogonal to the frame strands). In Figure 7, three displacement vectors (among all the displacement vectors possible) are represented as an example in schematic form. They are noted vî, v7

[0108] Each relative displacement vector has the action of sliding, relative to each other, columns of pixels so that the average curve f Mi(x^ merges with a pixel line (or the axis A of the image shown in FIG. 7 in relation to step E314). Each portion of the flattened average curve f MAi(x^ extends over a pixel column width of the mask (which corresponds to the dimension of a pixel of the mask in the direction of the pixel lines).

[0109] In other words, the displacement field u^x) is constant on each column of pixels. It is directed along the Y axis of the columns of pixels and depends on a position on the x axis: u^x) =

[0110] Thus defined, the displacement field u^x) corresponds, in practice, to a simple linear transformation: the columns of pixels are simply displaced (translated) without variation in surface area (in the XY plane) or volume when considering the direction along the main axis Z.

[0111] Furthermore, since the displacement field u^x) is directed in the sole direction of the pixel columns of the mask, it only has an effect in the relevant transverse XY plane. This ensures that each raw transverse image i1? of the set of raw transverse images i= 1' n is processed in step E31, independently.

[0112] This then makes it possible to process each raw cross-sectional image simultaneously, in parallel. This variant embodiment is illustrated in [Fig.8]. It offers the advantage of being very quick to execute.

[0113] As shown in Figures 6 and 7, the method continues at step E314. During this step, the section Si of the part 10 is subjected to the displacement field uj ( x ) to carry out the straightening.

[0114] Thus, in step E314 illustrated in figure 7, the displacement field u^x) determined previously is first applied to the pixels of the raw transverse image Y to obtain a straightened section of the part 10.

[0115] The application of the displacement field preferably corresponds to a translation (or an offset) and a linear interpolation.

[0116] For this, the displacement field m,(x) is decomposed into an integer part Un ( X ) and a real part Urî: u.( x ) = Uf. (x) + Ur^ The integer part is equal to the rounding down of the displacement field u^x); uy(x) -- [J •

[0117] Then, based on this displacement field, the rectified transverse image IriXY is obtained in two steps.

[0118] In a first step, we perform a translation (of whole pixels) to obtain a first rectified image ( x, y ): / (.( y ) = ( x, y + )

[0119] In a second step, the rectified image (x, y) is determined at from a vertical gradient of the first straightened image: 4 (x, y) = lu (x, y) • u^-

[0120] The determination of the vertical gradient A xy ) can be obtained by finite difference approximation *4(x, y +i)-z^y) oup ^ any other digital means.

[0121] For example, the determination of the gradient can be carried out by convolution with Sobel type operators. The rectified transverse image is then calculated in the following manner: / [( x, y ) = ( ( x, y + ^. ( x ) ) *KS ) • Ur / x) K* is a Sobel filter whose kernel is for example of type [ - 1, 0, + 1 ].

[0122] The displacement is thus composed of a translation (or an offset) and a one-dimensional interpolation of the gray levels which remains limited to half a pixel or less and is constant on each column of pixels. This then constitutes a very simple interpolation. This division of the displacement between an integer part and a non-integer displacement remainder is thus inexpensive in terms of computing resources.

[0123] Alternatively, this deformation may consist of a simple translation, or a simple interpolation.

[0124] As previously described in figures 6 and 8, step E31 is repeated on each raw transverse image

[0125] We thus have, at this stage of method 1, a set of rectified transverse images I\Xy^ i— 1 : n in which each rectified transverse image represents a straightened section of the part 10 (as illustrated by [Fig.7]).

[0126] As shown in Figure 5, the method continues with a step E32 of reconstructing a straightened three-dimensional image. This step E32 is carried out from the straightened transverse images.

[0127] Specifically, this step E32 consists of stacking the straightened transverse images I^Y 'cs one after the other along the main axis Z of the part. For this, each straightened transverse image flXY is identified by an axial coordinate which corresponds to the coordinate 1 of the corresponding section plane. The straightened three-dimensional image is then the volume formed by these assembled straightened transverse images. Similarly, in this three-dimensional image straightened / Çp, the straightened part is the volume formed by the straightened sections represented in the straightened transverse images I^y of the assembly.

[0128] The contribution and the interest of the method 1 are illustrated in particular with the support of a rectified two-dimensional image corresponding to a longitudinal section XZ of this rectified three-dimensional image. This longitudinal section is perpendicular to the Y axis of the XYZ reference of the part 10.

[0129] Figures 9A and 9B respectively represent an example of a longitudinal image obtained from the raw three-dimensional image and the corresponding rectified longitudinal image Irjj[Z obtained by applying the method according to the invention.

[0130] It is noted that the missing information in the longitudinal image / A obtained from the raw three-dimensional image is restored in the rectified longitudinal image fjxz- H it is thus possible to detect, by analysis of a single image, a weaving anomaly, where previously it was necessary to scroll through several images.

[0131] Thanks to the simple displacement produced by the displacement field, the texture of the part (i.e. the strands) are reconstructed without volume variations. We therefore obtain a rectified representation of homogeneous quality over its entirety.

[0132] Furthermore, the spatial resolution of the rectified longitudinal image l'jxz is sufficient to exploit sub-images of the rectified longitudinal image by enlarging them.

[0133] At the end of the straightening processing step E3, a straightened three-dimensional image is thus available which allows easy and reliable analysis of the weaving to be carried out.

[0134] When three local three-dimensional images I3D2, (see figure 3) are used to represent the part 10 in its entirety, method 1 is implemented for each of these local three-dimensional hü.b hD.?? images. We then obtain three rectified local images which, once concatenated, make it possible to obtain a global rectified three-dimensional image.

[0135] As previously, we choose a longitudinal section XZ associated with the image global rectified three-dimensional to illustrate the results obtained by method 1. Thus, Figures 10A, 10B, 10C and 10D each represent a longitudinal image straightened F obtained from a global rectified three-dimensional image. plus, each image represents a different 10 piece.

[0136] As can be seen in these figures 10A to 10D, each longitudinal image straightened / / ..., is formed from three straightened longitudinal images [- Y71 Ft' I -y? v each of these images being obtained from the rectified local three-dimensional image.

[0137] In all the figures, we observe good continuity between the straightened longitudinal images ^j^Z3- This is an advantage for visualizing in a single image the entire straightened part. This advantage is linked to the definition of the displacement field which provides an effect only in the transverse section planes and thus guarantees the independence of processing of the local three-dimensional images Iidi.

[0138] We also note the good repeatability of process 1 with respect to inter-blade differences. This is advantageous for a production application.

[0139] In an optimization mode, the step E32 of reconstructing the rectified three-dimensional image f3D may comprise an optional step of smoothing, along the main axis Z, of the rectified part 10. This smoothing step is optional.

[0140] This smoothing step may comprise determining a smoothing function for each XY plane of the straightened part 10.

[0141] This smoothing function is for example determined from a plurality of rectified transverse images of the rectified three-dimensional image. The determination of the smoothing function may for example be based on the determination of a weighting function of the displacement field u^x) associated with each rectified transverse image. This displacement field u^x) has for this purpose been stored using a storage means during step E313 of method 1.

[0142] This weighting function aims to adapt the amplitude of each displacement field u^x) so as to avoid discontinuities, along the main axis Z, between the different straightened sections S? forming the straightened part 10.

[0143] The weighting function can then be applied to each rectified transverse image I^y ■ Thus, the displacement field u^x) is updated.

[0144] The smoothing step is advantageously used to accelerate the calculation time of the processing method 1. Indeed, it makes it possible to limit the processing to a subset of raw transverse images of the plurality of raw transverse images IiXy. For example, this subset may contain one raw transverse image out of four of the plurality of raw transverse images. In this case, the smoothing step makes it possible to reduce the calculation time by four.

[0145] [Fig. 11] represents, in the form of a flowchart, a variant of implementation of method 1.

[0146] According to this variant, method 1 begins with a step EL

[0147] This step E1 consists of resizing the raw three-dimensional image from 5D of a predetermined three-dimensional region.

[0148] The resizing is performed first by extracting from the raw three-dimensional image the subset of voxels corresponding to the predetermined three-dimensional region.

[0149] Then, the resizing continues by generating a raw three-dimensional image from this selected subset. This generated raw three-dimensional image is called a "resized raw three-dimensional image".

[0150] The predetermined three-dimensional region is preferably an input parameter of the method 1. It is defined by dimension and position parameters in the raw three-dimensional image.

[0151] These parameters are for example chosen to include the voxels relating to the room 10 so as to limit the number of voxels not belonging to the room. For example, it is possible to limit them to the voxels complementary to the volume formed by the voxels relating to the room 10.

[0152] The resized raw three-dimensional image thus corresponds to a reduced file size since the number of voxels not relating to the part is reduced. It is easier to store and is processed more quickly by the processing method 1 described previously.

[0153] In another variant, the resizing step is not performed on the raw three-dimensional image, but on the transverse images. This step is then performed after step E2 and before step E3. Similar to step E1, this step consists of extracting a subset of pixels in the initial raw transverse image from a predetermined two-dimensional region of interest and then generating an image corresponding to the subset of extracted pixels. The predetermined two-dimensional region of interest may be an input parameter of the method 1 and preferably includes the pixels relating to the section of the part and the pixels complementary to the surface formed by these pixels relating to the section of the part.

[0154] Now, a second embodiment of the treatment method is described.

[0155] A first implementation variant is shown in [Fig.12], in the form of a flowchart.

[0156] According to this second mode of implementation, the processing method 2 aims to take into account the fact that the section Si of the part 10 is not elongated along the lines of pixels of the raw transverse image

[0157] For this, according to the first variant illustrated in figure 12, we first determine a projection frame (x', y') such that, in this projection frame, the section Si of the raw transverse image jb is transformed into an elongated section along an x Z^AI line of pixels of this raw cross-sectional image J^vv

[0158] Then, we define the average curve f^x') and the straightening of the raw cross-sectional part as a function of this projection reference.

[0159] Method 2 is identical to method 1 described previously except for steps E4, E5 and E6 which are new.

[0160] Step E4 aims to determine the projection reference frame defined previously. It is carried out between step E311 of generating the binary image and step E3123 of defining the parametric function.

[0161] This projection reference is preferably determined from the binary image B,. but can also be determined from the image C; of the column centers ym.

[0162] We consider here the case where it is determined from the binary image B^

[0163] We are looking for a rotation transformation relative to the center of the binary image Bj which makes it possible to find the alignment of the section Si with one of the lines of pixels of the binary image Bt. The rotation angle associated with the rotation transformation sought is the angle 0 referenced in [Fig. 13].

[0164] Specifically, this angle G is defined between a line of pixels of the binary image and a preferred direction d of the section Si of the part. The expression “preferred direction” designates the direction corresponding to the largest dimension of the section Si of the part 10 in the image considered.

[0165] According to a first variant of implementation of step E4, illustrated in FIG. 14A, the angle # is predetermined and not zero. Step E4 comprises a step E43 which consists of defining a system of reference change equations.

[0166] This system of equations for changing the reference frame is written, for example, as follows:

[0167] COS# sin# - sin#l / rxirxiirii where [ 1 is the position of the center of cos# J t .W + LJ the binary image.

[0168] In this embodiment variant, the angle # is constant for all images raw treated cross-sections.

[0169] Method 2 according to the second mode of implementation is advantageous for taking into account the relative orientation of the local three-dimensional images.

[0170] According to another variant of implementation of step E4, illustrated in FIG. 14B, the angle 0 is determined from the binary image, or from the image of the centers. As previously, we will consider the case where it is the binary image which is used.

[0171] First of all, in a first step E41, the pixels relating to the section of the part are extracted from the binary image. These pixels form a point cloud.

[0172] The geometric characteristics of this point cloud are then analyzed in a step E42 to identify its preferred direction.

[0173] There are several ways to carry out this step E42.

[0174] A first method is based on the use of a polynomial regression carried out from the point cloud.

[0175] According to a second method, a principal component analysis (or PCA) can be carried out along two perpendicular principal axes. This analysis then provides eigenvalues and associated eigenvectors allowing the point cloud to be modeled.

[0176] Then, according to this second method, it is a question of selecting, among the eigenvectors produced by the analysis by principal components, the eigenvector associated with the largest eigenvalue.

[0177] Then, the angle S is determined from the components of the selected eigenvector. Preferably, the determination of the angle θ is implemented by limiting the value range of this angle θ to the range [-45°, +45°].

[0178] Alternatively, a third method, consisting of calculating the geometric moment of inertia of the point cloud, then choosing as the preferred direction the direction which corresponds to the minimum moment of inertia can be used.

[0179] As a further variant, a fourth method consists of calculating the moment of inertia of the point cloud.

[0180] Among these methods, principal component analysis was preferred. This is in fact, in principle, close to the action sought for the displacement field. Step E4 then continues with step E43 described previously, which consists of defining the system of equations for changing the reference frame by taking into account the angle parameter 8 determined in steps E41 and E42.

[0181] This variant embodiment is suitable for a very general production method, in which the orientation of the part to be checked varies from one part to another: the treatment method 2 in fact has the capacity to adapt to each of the parts by taking into account its orientation.

[0182] Once the projection reference frame has been identified in step E4, the average curve in the projection reference frame is determined in step E5.

[0183] Step E5 is carried out between step E3122 of determining the image C, the column centers and step E3123 of determining the parametric function f-^x).

[0184] To do this, we apply the system of equations described previously to all the pixels of the image of the column centers C^x, y).

[0185] We then obtain an image of the centers C^x', y') in the projection frame. The parametric function is then defined from this image of the centers, in the projection frame.

[0186] Finally, after step E313 of determining the displacement field u^x') and before step E314 of rectification, the same system of equations is applied to the pixels of the raw transverse image Jbvv. A raw transverse image fiXY(x', y') is thus obtained in the same frame of reference as that which made it possible to determine the displacement field x').

[0187] A second implementation variant is shown in [Fig. 15].

[0188] According to a second variant of this second embodiment, illustrated in figure 15 in the form of a flowchart, the processing method 2 comprises, during the step E31 of processing each raw transverse image, a step E310 of image registration. This image registration step is optional.

[0189] This step E310 is then implemented before the step E311 of generating the binary image. It makes it possible to obtain from the raw transverse image J^vv, a registered raw transverse image IbXY ■ This registered raw transverse image IbXY is of the same dimension as the raw transverse image / ^y It has, like the raw transverse image, pixels arranged according to lines and columns of pixels, and comprises the section Si of the part 10.

[0190] The objective of step E310 is to align, in the registered raw transverse image fb.rvv, the columns of pixels according to the Y direction corresponding to the thickness of the part 10. Thus, the lines of pixels of the registered raw transverse image are oriented according to the X direction corresponding to the X frame direction of the reference frame of the part 10. As the section Si of the part 10 is elongated according to this X frame direction, this section Si is intrinsically represented in the registered raw transverse image Z^y in an elongated manner according to the lines of pixels of this registered raw transverse image jbr

[0191] The registration step E310 preferably begins with a step of determining the section Si of the part 10 in the raw transverse image.

[0192] Step E310 can then continue with a step of determining, from the determined section Si, the orientation of the columns of pixels of the raw transverse image fi relative to the orientation of the reinforcement (or thickness) Y axis of the part. ■* f^A 1 10.

[0193] Step E310 then comprises a step of resetting the section Si determined according to the determined orientation.

[0194] Finally, the recalibration step E310 makes it possible to position all the 3D volumes, i.e. all the raw three-dimensional images, in a common reference frame, that of the part 1. Advantageously, this step makes it possible to make the image analyses invariant to the positioning of the part 1 during acquisition.

[0195] Step E311 of generating the binary image is then carried out from the recalibrated raw transverse image. The other steps shown in [Fig. 15] are unchanged compared to the embodiment illustrated in Figures 6 and 7, and are not described again here.

[0196] A second aspect of the invention is a method 3 for controlling an aeronautical part. This method comprises the steps of method 1, 2 which has just been described and a step of controlling the part from the rectified three-dimensional image.

[0197] The control step comprises for example a step of visual analysis of three sets of images associated with the rectified three-dimensional image 7^.

[0198] Each set of images comprises images each corresponding to a section of this three-dimensional image rectified according to a section plane perpendicular to an axis of the XYZ reference frame of the part 10. The axis is different for each set of rectified images: for example, the axis is the Z axis for the first set, the Y axis for the second set and the X axis for the third set.

[0199] This control step is for example implemented manually or automatically.

[0200] A third aspect of the invention is the device 100 shown in [Fig. 16] and described previously. This device 100 is configured to execute the steps of the method 1, 2.

[0201] Finally, a fourth aspect of the invention is a computer program comprising instructions which, when the program is executed by the device, cause the latter to implement the steps of method 1, 2.

Claims

Claims

1. Method (1) for processing a three-dimensional image for su the control of a part (10) included in the three-dimensional image, the three-dimensional image being associated with a plurality of transverse images i — 1 : n) two-dimensional, each transverse image corresponding to a section (XY) of the three-dimensional image (l|D) according to a section plane perpendicular to a main axis (Z) of said part (10), the section planes being distinct, each transverse image (jb) comprising a section (Si) of the part (10), each transverse image being formed of a plurality of pixels (p(x, y)), X Ï»21L Y the method (1) comprising a rectification treatment (E3) comprising the following steps: - Processing (E31) of each transverse image ( / *vv) to obtain a plurality of rectified transverse images (Zp^y, i = 1 : ri), each rectified transverse image (^xy) comprising a rectified section (Sp of the part, said processing (E31) comprising, for each transverse image, the following steps: • Generation (E311), from the transverse image ^LXY^ concerned, of a binary image (^) to obtain a mask formed of a plurality of pixels corresponding to the pixels of the transverse image relating to the section (Si) of the part (10), the plurality of pixels of the mask (being arranged according to lines of pixels and columns of pixels, each pixel being defined by a line coordinate (x) and a column coordinate p), • Determination (E312, E3121, E3122), from the mask (^), of a surface representing the section (Si) of the part (10), said surface representing the section (Si) of the part (10) corresponding to

2. an average curve or a median curve of the section (Si) of the part (10), • Determination (E312, E3123) of a parametric function (f. ( x ), f ( x ) ) describing said surface, • Definition (E313) of a displacement field (Ui ( x ) ) constant on each column of pixels and dependent on the column coordinates of the pixels so as to apply a flattening transformation to the surface represented by the parametric function ( / ,- ( x ) ), • Straightening (E314) of the section (Si) of the part (10) in the transverse image (l^vv) concerned x ZA 1 by applying the determined displacement field (H, (x)) to the pixels of the transverse image ( / * v) concerned, to generate the rectified transverse image (^ïy) associated with the transverse image ( concerned, and Æ / A ï - Reconstruction (E32) of a rectified three-dimensional image (l|D) containing the part (10) rectified from the rectified transverse images (Z^jy) obtained from the raw transverse images (Z*vv)- x Z^AI Process (1) according to claim 1, in which the surface representing the section (Si) of the part (10) is determined by carrying out the following steps: - Definition (E3121) of a predetermined image (Pi) having the same dimension as the binary image, the predetermined image (P,) comprising a plurality of pixels (pp ( x, y ) ), each pixel (pp ( x, y ) ) having a predetermined value corresponding to the line coordinate (-^) of said associated pixel in the predetermined image - Assignment (E3122) of a predefined value to each pixel of the mask (Mi), said predefined value corresponding to the value of the pixel in the predetermined image (Pi) having the same row and column coordinates as the pixel of the mask concerned, and - For each column of pixels of the mask (Af,), determination of a center of said column from the predefined values of the pixels and the number of pixels of said column, and assignment of the determined center to each pixel of said column, the centers corresponding to a representation of said surface.

3. A processing method (1) according to claim 2, wherein the centers are determined from the result of a matrix product between a matrix representing the predetermined image and a matrix representing the binary image.

4. Processing method (1) according to one of claims 2 to 3, wherein the centers are barycenters, each barycenter being determined on the basis of the following expression: where: E,3(Xy) - x' y are respectively the column and row coordinates of each pixel of the mask, - Ip(x, y) is the matrix representing the predetermined image (Pi), and - B( x, y ) is the matrix representing the binary image (B^.

5. Processing method (1) according to one of claims 1 to 4, in which the parametric function is composed of a spline.

6. Processing method (1) according to one of claims 1 to 5, in which the application of the displacement field is broken down into a translation operation and / or an interpolation operation.

7. Processing method (1) according to one of claims 1 to 6, in which the plurality of rectified transverse images / Z j is "iXY} obtained simultaneously, by processing, in parallel, each transverse image z / 7 1 of the plurality of transverse images. "tXYj

8. Method (3) for controlling a part from a three-dimensional image comprising the steps of the processing method (1) according to one of claims 1 to 7 to obtain a rectified three-dimensional image and a step of controlling the part from the rectified three-dimensional image

9. Control method (3) according to claim 8, in which the part (10) is made of three-dimensional woven composite material.

10. Device (100) for processing a three-dimensional image For the inspection of a part (10) included in the three-dimensional image, the three-dimensional image being associated with a plurality of two-dimensional transverse images j — 1 ' n), each transverse image ( / *Yy) corresponding to a section of the three-dimensional image ( / ^) along a section plane (XY) perpendicular to a main axis (Z) of said part (10), the section planes (XY) being distinct, each transverse image comprising a section (¾ of the part (10), each transverse image being formed of a plurality of pixels, the device comprising a processor (110) configured to implement a rectification processing (E3) comprising: - Processing (E31) of each transverse image ( / ^Yy) to obtain a plurality of rectified transverse images ï = 1 • "))> each rectified transverse image ( / [^y) comprising a rectified section ($p of the part, said processing (E31) comprising, for each transverse image (f^vv), the following steps : • Generation (E311), from the transverse image concerned, of a binary image (2%) to obtain a mask (AfO formed of a plurality of pixels corresponding to the pixels of the transverse image relating to the section (Si) of the part (10), the plurality of pixels of the mask (being arranged according to lines of pixels and columns of pixels, each pixel being defined by a line coordinate (x) and a column coordinate (E), • Determination (E312, E3121, E3122), from the mask of a surface representing the section (Si) of the part (10), • Determination (E312, E3123) of a parametric function (f. ( x ), f ( x ) ) describing said surface, • Definition (E313) of a displacement field (¾ ( x ) ) constant on each column of pixels and dependent on the column coordinates of the pixels so as to apply a flattening transformation to the surface represented by the parametric function ( / ;(x) ), • Straightening (E314) of the section (Si) of the part (10) in the transverse image ( / ^y) concerned by applying the determined displacement field (iij(x)) to the pixels of the transverse image (J^y) concerned, to generate the straightened transverse image (Z / W) associated with the transverse image (T^y) concerned, and Reconstruction (E32) of a rectified three-dimensional image ( / ^) containing the rectified part (10) from the rectified transverse images

Images