Method for characterising a part

Abstract

A method of characterizing a part including obtaining an X-ray tomography image of the part and then a step of correlating the image with a reference wherein the correlation step includes searching among a predefined set of X-ray tomography image transformations for a transformation that minimizes the difference between the image and the reference in order to characterize the inside of the part.

Claims

1. A method of characterizing a part, the method comprising a step of obtaining an X-ray tomography image of the part and then a correlation step of correlating said image of the part with a reference comprising an image of a standard part differing from the part, wherein the correlation step comprises applying each of a predefined set of X-ray tomography image transformations to the image of the part to produce transformed images, determining respective differences between the transformed images and the reference, and searching among the predefined set of X-ray tomography image transformations for a transformation that minimizes a difference between the image of the part and the reference in order to characterize an inside of said part and determine a topological difference between the image of the part and the reference to find defects in the part.

2. A method according to claim 1 for characterizing a part, using a continuous parameterization of said set of transformations.

3. A method according to claim 1 for characterizing a part, wherein said set of transformations includes at least one set of continuous displacements, at least one set of alterations of brightness and of contrast, or at least one set of scale changes.

4. A method according to claim 1, for characterizing a part, further comprising determining whether the part is acceptable by using an extracted transformation.

5. A method according to claim 1, for characterizing a part, wherein said set of transformations comprises at least transformations corresponding to modifications of at least one parameter of a model of the part.

6. A method according to claim 1, for characterizing a part, wherein the reference comprises a virtual part constructed from a computer assisted design model.

7. A method according to claim 1, for characterizing a part, including modifying a parameterization of a computer assisted design model of the part by using a transformation identified at the end of the search.

8. A method according to claim 1, for characterizing a part, wherein said set of transformations comprises at least a class of transformations that conserve topology.

9. A method according to claim 1, for characterizing a part, wherein the reference comprises an explicit representation of boundaries of an element of the part.

10. A method according to claim 1, for characterizing a part, including segmenting the image of the part by using a transformation identified at the end of the search.

11. A method according to claim 1, for characterizing a part, wherein the reference comprises a representation of an elementary pattern.

12. A method according to claim 1, for characterizing a part, comprising identifying at least one of strands, fibers and elementary patterns in the image of the part.

13. A method according to claim 1, for characterizing a part, wherein the part is a composite material part.

14. A method according to claim 1, for characterizing a part, wherein the part is an aircraft turbojet blade.

15. A method according to claim 1, comprising: identifying a specific component within the image of the part; situating the specific component from a reference image in the image of the part; and using the set of transformations to deform the specific component in the reference image to match the image of the part.

16. The method according to claim 1, wherein: the set of transformations comprises a plurality of transformations; and the searching comprises: reconstructing an image using each transformation of the set of transformations to generate the set of transformed images, and finding a difference between the reference and each of the transformed images.

17. The method according to claim 1, wherein the set of transformations comprises a plurality of transformations.

Description

LIST OF FIGURES

(1) The description of the invention is continued below with reference to the accompanying figures.

(2) FIG. 1 shows an implementation of the invention. FIGS. 2 and 3 show examples of this implementation.

(3) FIG. 4 shows a second implementation of the invention. FIG. 5 illustrates this implementation.

(4) FIG. 6 shows a third implementation of the invention.

DESCRIPTION OF IMPLEMENTATIONS

(5) In an implementation as shown in FIG. 1, use is made of a real reference image 20, such as the image of a part that is used as a standard or a template. A part 10, e.g. a composite material part, is reconstructed in a raw tomography image of the part 10, made up from a quantity of tomographic data 100 (i.e. a number of projections) that may be small. The reconstruction is performed using a set of transformations 30 that are considered to be realistic. The computation is based on a step of searching for and identifying the transformation T* (reference 40) in the set 30. This identification is performed by searching for the minimum (optimization 200) as described in the introduction. Simultaneously, the associated topological difference (reference 50) is determined.

(6) In a variant, the differences between the part under study and the standard part are found and identified effectively by using the topological difference (reference 50) during a step 300. For example, as shown in FIG. 2, comparing tomographs for a composite material part before and after testing under load shows up very clearly the presence of mesocracks.

(7) It is also possible to establish a correlation between two different samples, and if they are parts made of composite material, this can reveal differences from a point of view weaving between the two samples. This is shown in FIG. 3.

(8) In general manner, it is thus possible to perform non-destructive tests (NDT) on composite material parts, e.g. turbojet blades. The technique described leads to savings in time for inspecting, and acquiring and storing data. Thus, by way of example, during a step 350, it is possible merely on the basis of the transformation T*, to decide whether the part should be retained or rejected.

(9) In another implementation, shown in FIG. 4, two images are put into correspondence, one a real image and the other an image referred to as a virtual image, e.g. an image 20 made up from a computer assisted design (CAD) model of the part.

(10) Under such circumstances, the parameters 30 of the CAD model may themselves comprise a specific transformation class T. Thus, tomographic data can be used on the basis of the image of the CAD model by writing the tomographic image directly in a description language suitable for dialog with the CAD design team of the part.

(11) The dialog then consists in providing, in the form of the CAD model, a good predetermination of the solution for assisting in constructing the tomographic image (step 200) on the basis of the tomographic data. In return, the image as constructed in this way then makes it possible, in a step 400, to correct the CAD model by means of the parameters of the identified transformation T* so that it is as close as possible to the part actually made.

(12) The method is performed until the algorithm used converges or becomes stationary, e.g. in a simple context of adjusted gray levels.

(13) Defects of orientation or of alignment can affect the response of the complete composite structure, and an initial adjustment as proposed is a good way of improving and validating a CAD model taking account of such imperfections.

(14) 3D models can be generated that are made discrete in the form of individual voxels or that are represented by a parametric model, or a computer assisted design (CAD) model, based on a priori knowledge about the woven array of composite fiber reinforcing materials. It is thus possible to correlate the image of a part and an image derived from a mold and to modify the input parameters of the model, i.e. the directions of the strands and also their dimensions. By way of example, FIG. 5 shows an image obtained by tomography and an image of a model.

(15) In a variant, if the paths followed by the yarns are not included in the reference image obtained by a CAD model, the paths of the strands are determined directly from the tomographic image, e.g. by using a tracking algorithm, which is provided with the results of the correlation with the reference image obtained by a CAD model.

(16) In a variant shown in FIG. 6, images are segmented. Segmentation 500 consists in identifying specific components within an image, e.g. strands or fibers or elementary patterns (e.g. a phase modulated periodic pattern), that might possibly be contained in a database constituting a dictionary. If a theoretical representation of the article to be identified is available (looked-for topology 20), it is then possible to correlate 3D images of this article with data 100 obtained by tomography. The transformations 30 used are performed with imposed topology, and make it possible to conserve the topology of the reference article in robust manner.

(17) Using the identified transformation T*, the defined component in the reference image can thus be situated in the image and can be deformed in order to match the real image.

(18) Thus, if it is desired to find a closed curve, it suffices to start with an ideal image of a closed line such as a perfect circle and then allow it to vary progressively towards the line as is present in the image of the medium.

(19) This approach is more robust than the usual thresholding and segmenting techniques that do not automatically preserve the correct topology for the looked-for article. Thus, with these techniques, missing points in a curve that ought to be closed are obtained, or a thick curve is obtained when it ought to be fine.

(20) The segmentation as performed in the described method, i.e. automatically on the basis of a previously defined topological element, serves to minimize intermediate steps of image filtering where the contributions of noise, bias, and measurements are not always easily determined, and thus where information is easily degraded by such filtering.

(21) The invention is not limited to the implementations described, but extends to any variant within the ambit of the scope of the claims.