On the Design of Acousto-Ultrasonics – Pattern-Recognition Classifiers for the Identification of Material Response

This paper deals with the design procedure of computer-based Pattern Recognition (PR) Classifiers, when the Acousto-Ultrasonics (AU) technique is employed for the identification of material response states. For this purpose, material specimens presenting different mechanical response states are processed to retrieve AU signals in the form of digitalized records. The retrieved AU data are, then, grouped in distinct classes, each pertaining to a known response state of the material under consideration. These AU data are, subsequently, identified by pattern vectors, the components of which represent values of characteristic features in certain domains of description. PR-classifiers built from these pattern vectors are subsequently used to identify unknown response states of the material by classifying their pertaining AU signals as belonging to one of the predefined classes. In the paper, AU-experimental work in conjunction with PR-computational results are presented to illustrate the design procedure and testing of PR-classifiers for the purpose of identifying various low-energy repeated-impact states of a class of polymeric material; namely, Polyvinylchloride (PVC).