Pot-Core Hall probe with multivariate empirical mode decomposition based on spatial filtering for quantitative pulsed eddy current measurement of subsurface corrosion

• A pot-core Hall probe for PECT overcomes SNR and lift-off limits of conventional air-core probes. • Spatial 2D-EMD is validated for adaptive filtering of PECT C-scans to isolate corrosion signatures. • Optimized hardware plus adaptive spatial processing enables accurate quantitative inversion of corrosion volume. • Backside inspection detectability improves markedly, with over three-fold CNR gain using the pot-core probe. • MEMD-derived features yield robust corrosion quantification, achieving R 2 = 0.90 in volume prediction. Accurate quantification of subsurface corrosion remains a persistent challenge in pulsed eddy current testing (PECT), particularly when conventional probes suffer from weak field coupling, low signal-to-noise ratio (SNR), and strong sensitivity to lift-off variations. This study presents a high-fidelity measurement framework that integrates a purpose-designed pot-core Hall sensor with an adaptive spatial signal processing strategy to enable reliable quantitative characterization of hidden corrosion. A three-dimensional finite element model is first developed to examine the transient eddy current diffusion and to clarify how the ferrite pot-core architecture concentrates magnetic flux and enhances probe–specimen interaction. The fabricated pot-core probe is then validated experimentally on an Al2024 multilayer specimen containing nine artificial corrosions of varying diameters and depths. Compared with a conventional air-core probe, the proposed design achieves more than a three-fold improvement in mean contrast-to-noise ratio during backside inspections, demonstrating substantially improved robustness against lift-off effects. To further enhance image interpretability, a multivariate empirical mode decomposition (MEMD) filter is applied to the raw C-scan maps, adaptively separating localized corrosion signatures from high-frequency noise and nonlinear background drift. The refined images enable extraction of physically meaningful features, which are used to develop simple multivariate regression models for corrosion quantification. The proposed framework achieves a high prediction accuracy for corrosion volume (R 2 = 0.90 on validation data), confirming the synergy between the optimized sensor architecture and adaptive spatial decomposition. The results demonstrate a practical and scalable approach for high-accuracy PECT measurement in demanding inspection scenarios.