<scp>GMR</scp> ‐Based Eddy‐Current Sensing and Embedded <scp>TinyML</scp> for Advanced Crack‐Shape Characterization

ABSTRACT A combined computational and experimental approach was used to characterize crack shapes in flawed specimens. The first step involved a three‐dimensional finite‐element method based on the (A, V–A) formulation to analyze field variations in cracked conductive materials and to evaluate the influence of defect geometry on the eddy‐current response. This numerical model enabled the determination of crack shapes. The study also employed a giant magnetoresistance (GMR) sensor to measure signals from different crack forms using a GMR‐based eddy‐current (EC) probe. The model was validated experimentally through a prototype unit, and measurements were performed on aluminum reference standards containing various crack types. Furthermore, a TinyML model was developed using the Edge Impulse platform to automatically classify crack shapes according to relevant standards. Using the GMR‐based EC probe, the system achieved a mean accuracy of 98%, demonstrating the feasibility of the method. A key advantage of this approach is the rapid and efficient development of embedded machine‐learning models enabled by the open‐source platform. The approach offers a cost‐efficient solution for industrial NDT, with future improvements focused on expanding the dataset and validating system performance in real operating environments.