3D shape identification of parallelepiped flaw by means of biaxial MFLT using neural network

ABE Masataka¹, BIWA Shiro², and MATSUMOTO Eiji³

1. Graduate School of Energy Science, Kyoto University, Kyoto, 606-8501, Japan (m.abe@ax7.ecs.kyoto-u.ac.jp)

2. Graduate School of Engineering, Kyoto University, Kyoto, 606-8501, Japan (biwa@kuaero.kyoto-u.ac.jp)

3. Graduate School of Energy Science, Kyoto University, Kyoto, 606-8501, Japan (matumoto@energy.kyoto-u.ac.jp)

Abstract: In this paper, we attempt to evaluate the three-dimensional shape of a parallelepiped flaw and  identify its location, i.e. the horizontal position and the located surface, by means of biaxial Magnetic Flux Leakage Testing (MFLT), employing a Neural Network (NN). The specimen is a magnetic material (SS400) subjected to a magnetic field, and the magnetic flux in the specimen leaks near the flaw. We measure the biaxial Magnetic Flux Leakage (MFL), i.e., the tangential and the normal components of the MFL, along a line parallel to the specimen's surface. We then approximate the measured biaxial MFL distributions by means of elementary functions with a small number of coefficients. The approximation coefficients are extracted as  Characteristic Quantities (CQs) of the MFL distribution. The horizontal position of the flaw along the measurement line is characterized by some of these CQs. NN is used to infer the cross section of the flaw, i.e., the width, depth, and located surface of the CQs. By repeating a similar process along several measurement lines parallel to the specimen's surface, we can identify the three-dimensional shape of the flaw, including its location. The NN, trained with several known flaws, was found to be able to evaluate the three-dimensional shape and location of a parallelepiped flaw with a high level of accuracy.

Keyword: NDT; MFLT; neural network; flaw detection; magnetic sensor