Elman neural networks for characterizing voids in welded strips: a study

Within the framework of aging materials inspection, one of the most important aspects regarding defects detection in metal welded strips. In this context, it is important to plan a method able to distinguish the presence or absence of defects within welds as well as a robust procedure able to characterize the defect itself. In this paper, an innovative solution that exploits a rotating magnetic field is presented. This approach has been carried out by a finite element model. Within this framework, it is necessary to consider techniques able to offer advantages in terms of sensibility of analysis, strong reliability, speed of carrying out, low costs: its implementation can be a useful support for inspectors. To this aim, it is necessary to solve inverse problems which are mostly ill-posed; in this case, the main problems consist on both the accurate formulation of the direct problem and the correct regularization of the inverse electromagnetic problem. We propose a heuristic inversion, regularizing the problem by the use of an Elman network. Experimental results are obtained using a database created through numerical modeling, confirming the effectiveness of the proposed methodology.