文摘
A common target of clustering in acoustic emission (AE) non-destructive inspection technique (NDT) is to distinguish between the sources of different origin and to get a deeper insight into the interrelation between the underlying processes such as plastic deformation, crack initiation, corrosion cracking, etc. The major drawback of the most popular conventional schemes such as k-means and fuzzy c-means is that they are iterative in nature, which hinders their real-time applications. Inspired by the sequential k-means procedure, i.e. a non-iterative variant of the classic k-means, we present a novel classification technique designed for real-time applications. The proposed approach is ¡°non-supervised¡±, i.e., both the number of clusters and their elements are inferred from the data distribution in a multi-dimensional metric space. In its present form the approach is capable to adopt various dissimilarity measures to compare AE power spectral densities. A series of tests on different probing datasets has been performed to prove the efficiency of the proposed approach.