Involved people: Arnaldo de Albuquerque Araujo, Marcos C Andrade, Gilles Bertrand, Michel Couprie, Laurent Letellier (LETI, CEA), Régis Vinciguerra (ESIEE student).
In the nuclear power stations intended for the production of electricity, the main material used is the uranium oxide, conditioned in the shape of ceramics pellets. These pellets are obtained by compression of a powder made up of oxide grains, ceramics thus presents a heterogeneous structure which can be highlighted by microscopy: see fig. 14(a) a portion of a cross-section of one of these pellets (the size of a grain is about 10 µm). The behavior of the material during the operation of the reactor depends, among other factors, on the size of the oxide grains and the porosity of the pellets (size of the joints, presence of cavities). To study these parameters which are 3-dimensional by nature, some methods were developed which make it possible to start from a measurement of two-dimensional statistical characteristics, that can be obtained from images of cross-sections (fig. 14(a)).
The principal problem is for us to divide the image into its various grains: we call this operation a segmentation . Although realized easily by the human brain, the segmentation of an image can pose difficult problems, often due to the presence of noise and the discretization of the data. We contributed to the analysis of such images by proposing new methods of segmentation based on concepts of topology for grayscale images [ BEC97, CBB99], and by implementing efficiently some methods from the framework of mathematical morphology [ CB97]. A "good" segmentation of the initial image, obtained thanks to one of our methods, is presented fig. 14(b). We recieved a financial support from the LETI-CEA for a study related to the analysis of such images.
 |
 |
| (a) | (b) |