Automatic flaw detection in NDE signals using a panel of neural networks

Abstract

The magnetic flux leakage (MFL) method is widely used for the assessment of the condition of off-shore cables, gas pipes and aerial tramway cables. At Empa, a measuring device was developed to use the physical effect of MFL for large diameter steel cables as used in civil engineering structures. It is considered that the automatization of the flaw detection task would offer great advantages, both from an economic point of view as well as in terms of reliability. In this work the application of a feature extraction model based on an analytical approximation of the MFL intensity is presented. Parameters that are closely related to the size and position of a defect are used as fitting variables in the solution of the inverse problem that links the recorded MFL intensity to the flaw that generates it. Using these significant parameters a committee of neural networks is trained in order to obtain a robust flaw detection algorithm.