Image analysis in industrial radiography

Non-destructive testing in nuclear power plants remains a major EDF objective for the coming decades. To facilitate diagnosis, the expert must be provided with elaborate decision-making aids: contrasted images, noise-free signals, pertinent parameters, ''meaningful'' images. In the field of industrial radiography, the inspector's offer of a portable system for digitalization and subsequent processing of radiographs (ENTRAIGUES) is an improvement in the inspection of primary circuit nozzles. Three major directions were followed: - improvement of images and localization of flaws (2D approach); techniques such as Markov modelling were evaluated and tested, - development of a system which can be transported on site, for digitalization, processing and subsequent archiving on inspection radiographs, known as ENTRAIGUES, - development of a program for aid in analysis of digitized radiographs (''bread-board'' version), offering an ergonomic interface and push-button processing, which is the software component in ENTRAIGUES and uses sophisticated methods: contrast enhancement, background flattening, segmentation. An other objective is to reconstruct a three-dimensional volume on the basis of a few radiographs taken at different incidences and to estimate the flaw orientation within a piece understudy. This information makes sense to experts, with regards to the deterioration rate of the flaw; the equipment concerned includes the formed bends in the primary coolant nozzles. This reconstruction problem is ill-posed and a solution can be obtained by introducing a priori information on the solution. The first step of our algorithm is a classical iterative reconstruction A.R.T. type method (Algebraic Reconstruction Techniques) which provides a rough volumic reconstructed tridimensional zone containing the flaw. Then, on this reconstructed zone, we apply a Bayesian restoration method introducing a Markov Random Field (MRF) modelling. Conclusive results have been obtained. (author). 2 figs., 4 refs