SVM classification of thickness and lift-off using transient eddy current oscillation method

Abstract

Wall-thinning in stainless steel structures due to corrosion and surface erosion produced by the harsh operating conditions and by the flowing commodity origins pipeline failures. Thus, the detection of a change in local wall thickness of pipelines is extremely important for accident reduction. A new nondestructive testing (NDT) method called transient eddy current oscillations (TECO) is used in this paper to detect a change in the thickness of stainless steel covered with insulation. This method uses a probe which consists of a capacitor and an induction coil (excitation coil) discharging in a free oscillation fashion, while the resultant magnetic field is detected by using a Hall-effect sensor. Stainless steel test specimens of different thicknesses were prepared to test the feasibility of the method. Several thicknesses of plastic spacers were used to simulate the insulation on the steel plates. Different features were extracted from the response voltage to detect the thickness of the test specimen and separation of lift-off. To improve accuracy, a model-based support vector machine (SVM) analysis was applied on the experimental data. By incorporating several features at the same time, this technique avoids and surpasses the limitation of the traditional NDT methods used for prediction of wall-thinning.