Essentially qualitative in its infancy, X-ray tomography today meets precise quantitative needs aimed at better identifying the defects detected in the imaged parts. Based on modeling the behavior of materials and predicting their lifetime, tomography is part of an industrial strategy that has evolved from “zero defects” to “acceptable defects”. Analyzing a defect, with a view to its possible acceptance, means precisely knowing its nature, its dimensions and its location within the material in order to qualify its harmfulness.

Once the volume imaging of an industrial part has been obtained using the X-ray tomography technique, the challenge, in the context of a better knowledge of material health, is often to determine the spatial distributions, in shape and in size of defects of different origins. This information is fundamental when it comes to understanding the damage mechanisms that can appear under mechanical, chemical or thermal stresses, but also to feed or guide the construction of an appropriate model. The main objective of this postdoctoral study is the establishment of a three-dimensional learning analysis method of defects obtained by X-ray tomography. The defects concerned are those resulting from manufacturing processes for metal alloys such as additive manufacturing or foundry.