Method for determining the presence of one defect on a tomographic image of a three-dimensional part by breaking the image down into principal components

1 . Method for determining the presence of at least one defect in a three-dimensional part coming from a production line according to a reference base, comprising the following steps:

a) assembling a set of defect-free parts comprising a predefined number of parts, considered to be without defects and considered to be representative of the parts out of the production line,

b) for each defect-free part of the set, the following steps are carried out:

i. acquiring an image of the defect-free part,

ii. segmenting the image of the defect-free part,

iii. registering the segmented image of the defect-free part with respect to an image in three dimensions of the part coming from computer-aided design,

iv. dividing the registered segmented image of the defect-free part into at least two subdomain images, and

v. for each of the subdomain images,

(1). registering the subdomain image on the three-dimensional image obtained by computer aided design, and

(2). recording the registered subdomain images in a tensor,

c) for each subdomain,

i. carrying out the union of the tensors of the registered subdomain images across the defect-free parts of the set in a rewritten tensor, then

ii. determining principal modes of the rewritten tensor via a decomposition into principal components, and

iii. recording the principal modes obtained, for the subdomain, in the reference base,

d) providing a part to test for defects from the production line, different from the set of defect-free parts,

e) acquiring an image of the tested part,

f) segmenting the acquired image of the tested part,

g) registering the segmented image of the tested part with respect to a representation in three dimensions of the part coming from computer aided design,

h) dividing the registered segmented image of the tested part into at least two subdomain images, the subdomains associated with the registered image being determined similarly to the determination of subdomains involved in determining the reference base, and

i) for each subdomain image of the tested part, the following substeps are carried out:

i. registering the subdomain image of the tested part on the three-dimensional image obtained by computer-aided design,

ii. projecting the registered image of the subdomain of the tested part onto a modal base associated with this subdomain, the modal base comprising the principal modes of the reference base for the subdomain,

iii. generating a synthetic image of the subdomain from the projection,

iv. obtaining a residual image by subtracting the synthetic image of the subdomain and the subdomain image, in order to eliminate the defect-free elements off the three-dimensional image, and

v. determining that a defect is present in the subdomain of the tested part in particular by comparing the value of each voxel of the residual image of the subdomain of the tested part to a predetermined threshold.

2 . Determination method according to claim 1 , wherein each segmented image of each defect-free part is registered to eliminate possible offsets.

3 . Determination method according to claim 1 , wherein each subdomain image of each defect-free part is registered with respect to the corresponding subdomain image of the reference base to eliminate the possible offsets.

4 . Determination method according to claim 1 , wherein the image of each defect-free part is acquired by tomography, by temporal imaging, by radiography or by optical imaging.

5 . Determination method according to claim 1 , wherein the image of the tested part is acquired by tomography, by temporal imaging, by radiography or by optical imaging.