Modeling increases analysis reliability: the forward model computes the synthetic eddy current signal of a defect from its geometrical description. MASCOTTE is the forward model used in this application. The inversion process defines the defect from the eddy current signal: the result is the best correlation between the forward model signal and the eddy current signal.

We present eddy current signal inversion for axisymetric defects using bobbin coil. The MASCOTTE code, with its simplify meshing, allows fast simulation of eddy current synthetic signal. Non linear inversion implies the use of knowledge to be included in the analysis procedures (e.g. signal calibration, depth evaluation, ...). We demonstrate that optimization based on global models is necessary to converge to the best solution. Enhanced Simulated Annealing (ESA) algorithm converges to the best coherence between the forward model signal and eddy current signal, whatever the 2D defect configuration is required. Comparisons between ESA algorithm, simplex algorithm and genetic algorithm will be thoroughly described.