Many physical phenomena in the interaction of an ultrasonic beam with defects make difficult the interpretation of testing data: superimposition of echoes travelling along different wavepaths between transducers and defects (specular reflection, diffraction, mode conversion), transducer diffraction effects, echoes generated by the piece under test, multiple reflection... We present the model Méphisto for predicting ultrasonic images (A-scan, B-scan and C-scan) in testing configurations. Hypotheses for this model allow short time computation. For a best accuracy between predicted and measured images, Méphisto can use results of Champ-Sons, a model predicting the broadband field radiated into a piece by an arbitrary transducer.

The characterization of defects consists in the extraction of the maximum available information about the defect from ultrasonic images. We present two specific inverse methods. One was developed to discriminate planar (cracks) and volumetric defects using their A-scan signatures. The other one is used to characterize Outer Surface Defects (OSD), i.e. defects located in the vicinity of the outer surface. It is based on a comparison between an experimental image obtained from the OSD control and the corresponding image computed with the Méphisto model. With an initial hypothesis on the kind of defect (supposed to be a planar axially oriented defect) a criterion of likelihood together with a minimization are applied to estimate the geometrical parameters of the defect (size, orientation and position). This whole procedure automatically converges to a solution for the set of parameters offering the best match between the simulated and the experimental segmented B-scans. Using the segmented B-scan rather than the original B-scan is an original approach presenting several advantages: it concentrates the useful information and avoids a ill-posed problem, reduce operational requirements. Actual cases are treated to validate the presented inversion procedures.