Relative performance of clustering-based neural network and statistical pattern recognition models for nondestructive damage detection

Abstract

The objective of this paper is to compare and contrast the capabilities of neural networks and statistical pattern recognition to localize damage in three-dimensional structures. A theory of damage localization, which yields information on the location of the damage directly from changes in mode shapes, is formulated. Next, the application of statistical pattern recognition and neural networks for nondestructive damage detection (NDD) is established. Expressions for classification using linear discriminant functions and a two-stage supervised clustering-based neural network are generated. Damage localization is applied to a finite-element model (FEM) of a structure which contains simulated damage at various locations. A set of criteria for comparing and contrasting statistical pattern recognition and neural network models is then established. Finally, the evaluation of the two models is carried out using the established criteria.