Efficient Numerical Approach and Experimental Validation for MAPOD in Eddy Current Nondestructive Testing

Abstract

Model-assisted probability-of-detection (MAPOD) studies for eddy current (EC) nondestructive testing (NDT) require a large number of flaw evaluations across a wide range of defect dimensions, making computational efficiency a critical consideration. This article presents an efficient numerical framework for characterizing and simulating the response of surface-breaking flaws detected using a reflection differential split-D EC probe. A detailed finite-element model was developed in COMSOL Multiphysics and optimized through the implementation of impedance boundary conditions (IBCs), symmetric meshing strategies, and symmetry plane model reductions to achieve an effective balance between simulation accuracy and computational cost. Single-parameter and multiparameter sweeps were performed to investigate the influence of crack dimensions and liftoff on the differential probe response. The resulting simulation database was used to construct a surrogate model capable of predicting the peak differential impedance response for arbitrary flaw dimensions within the investigated parameter space without requiring direct finite-element evaluation. The surrogate formulation captures the dependence of the response on flaw depth, length, width, and liftoff and achieves agreement with the numerical simulation data with a correlation coefficient of $R^{2} =97.26$ %. To validate the proposed approach, raster-scan measurements were performed on a set of titanium specimens containing electrical discharge machining (EDM) notches of varying lengths using an automated EC scanning system. The measured responses exhibited trends consistent with the simulation results. Furthermore, the experimental dataset was used to establish a transfer function relating the surrogate model predictions to measured responses, resulting in a correlation coefficient of $R^{2} = 94.95\%$ . The proposed framework provides an efficient pathway for generating large datasets required in MAPOD studies while maintaining strong agreement with experimental observations.