A novel feature extraction method of eddy current testing for defect detection based on machine learning

Y Tao, W Yin and A Peyton 

In eddy current testing, the trajectory of the impedance data due to a defect is presented as a Lissajous curve (LC) in the complex plane. This paper proposes a novel analytical model for describing an LC. Furthermore, a new feature extraction method is implemented, which automatically computes four geometric features (amplitude, width, angle and symmetry) from Lissajous figures. In addition, six machine learning-based classifiers are used for automatic defect identification based on these features. High detection rates are achieved for both the simulated and experimental data, which demonstrates the flexibility of the analytical model and the validity of the methodology.