A single-frequency approach for crack dimension estimation using ECT based on classification and regression fusion

This paper presents a new method to classify defective materials inspected by non-destructive testing (NDT) based on eddy currents (EC), including estimating the dimensions of small superficial and buried defects. The data processing uses lightweight machine learning steps that include a new kind of feature extraction, data augmentation, and a fusion of defect classification and defect dimension estimation. There is also a theoretical discussion on the suitability of the fusion stage. Several experiments were performed, acquiring a dataset of eddy current signals by means of an advanced EC probe that includes two pairs of exciting coils, generating two orthogonal magnetic fields, and three magnetic field sensors. Only a single excitation frequency was required, thus making it simpler to set up experiments, while reducing the measurement time and the computational complexity of the process. Several aluminum specimens (up to 30) from different kinds of materials were evaluated, demonstrating the superiority of the proposed method compared to previous works. There were improvements in estimating the defects’ depth and height, yielding errors below 0.13 mm and 0.19 mm respectively.