Classification of uncertain data: An application in nondestructive testing

Abstract

The classification of data with dynamically changing uncertainty characteristics is a problem for many practical applications. As an example in the field of nondestructive testing (NDT), magnetic flux leakage (MFL) measurements are used to inspect pipelines. The data is analyzed manually afterwards. In this paper we use a framework for handling uncertainties called Trust Management and a extended fuzzy rule based classifier to identify different installations within pipelines by MFL-data. The results show a classification performance of over 90% with an additional, reliable measure for the trustworthiness of every single classification result. © 2012 Springer-Verlag Berlin Heidelberg.