Nondestructive evaluation of loose assemblies using multi-frequency eddy currents and artificial neural networks

Abstract

This paper considers the problem of the evaluation of metallic assemblies in an aeronautical context, by means of a non-invasive method. The problems lies in the estimation of the distance separating two aluminum plates representative of a loose assembly (up to 300 µm), the top plate being possibly of unknown thickness ranging from 1 to 8 mm. To do so, the eddy current (EC) method is chosen, because it allows non-contact evaluation of conducting media to be carried out, which is sensitive to electrical conductivity changes in the part under evaluation, and hence to the presence of an air gap between parts. The problem falls into the category of evaluation of a multilayered conductive structure starting from EC data, which is an ill-posed problem. In order to bypass these difficulties, as well as to deal with the uncertainties that may be introduced by the experimental set-up, a ‘non-model’ approach is implemented by means of an artificial neural network (ANN). The latter is elaborated in a statistical learning approach starting from the experimental EC data provided by a ferrite cored coil EC probe used to investigate an assembly mockup of adjustable configuration. Moreover, in order to build a learning database allowing a robust and accurate ANN to be elaborated, as well as to deal with assemblies of unknown thicknesses, we consider EC data obtained at different frequencies chosen in an adjusted frequency bandwidth, experimentally determined so as to optimize the sensitivity toward the presence of an air gap between parts. The implementation of the proposed approach for distances between parts ranging from 60 to 300 µm provided estimated root mean square errors ranging from 7 μm up to 50 µm for the estimation of the distance between parts, and ranging from 20 µm up to 1.4 mm for the estimation of the top plates, ranging from 1 to 8 mm, respectively.