Sizing of natural crack using multi-output support vector regression method from multi-frequency eddy current testing signals

Abstract

In the nondestructive evaluation for components of key equipment, sizing of natural crack is important in order to guarantee both the safety and efficient operation for large mechanical systems. Natural cracks have complex boundary and there may be electric current flowing through crack faces. If a simple model of artificial notch is used to simulate it, errors often occur in crack depth reconstruction from eddy current testing (ECT) signals. However, if a complex crack conductivity model is used, quantitative evaluation of natural crack will be transformed into a multivariable nonlinear optimization problem and the solution is difficult. In this paper, based on the relationship between crack parameters and features of multi-frequency ECT signals, a multi-output support vector regression algorithm using domain decomposition for parameters was proposed. The algorithm realized the quantitative evaluation of multiple parameters of crack in turn. Numerical examples with simulated and measured ECT signals were presented to verify the efficiency of the proposed strategy.