Eddy current conductive coating layer assessment on conductive substrate: a machine learning approach

The accuracy in evaluating conductive layer coating thicknesses over conductive materials has recently been advanced with the potential application of apparent eddy current conductivity (AECC) spectroscopy. This approach minimises the sensitivity to variation in lift-off distance between the used eddy current coils and tested samples while capturing the actual conductivity profile with accuracy. This investigation further explores the potential integration of AECC spectroscopy with emerging machine learning (ML) capabilities in a theoretical-based supervised learning approach using artificial neural networks (ANNs). The complex spatial integral transform of the eddy current signal effectively filters out the information related to abrupt changes in the conductivity profile of step-like multi-layer structures, which renders the application of ANNs unfeasible for such structures. However, in the case of rectangular conductivity profiles covered in this study, it not only demonstrates the accuracy in estimating AECC spectrums over a range of different coating and substrate conductivity combinations and their associated coating thicknesses but also the retrieval of these unknown input parameters from measured AECC spectrums. The implementation of the ANN inverse model to experimentally measured AECC spectrums resulted in a maximum uncertainty of 3.37% in the conductive layer thickness estimation within a practical lift-off range.