Experimental Characterization of Ultrasonic Phenomena by a Neural-Like Learning System

Abstract

This paper describes a novel approach for analyzing ultrasonic signals to permit an experimental determination of the relations between elastic wave phenomena and the properties of a source of sound in a material. It is demonstrated that an adaptive learning system comprising an associative memory can be used to map source and waveform data and vice versa with the auto-and cross-correlation portions of the associative memory. Experiments are described wh ich utilize such an adaptive system, running on a laboratory minicomputer, to process the data from a transient ultrasonic pulse in a plate specimen. In the learning procedure, the system leams from experimental pattern vectors, wh ich are formed from the ultrasonic waveforms and, in this paper, encoded information about the source. The source characteristics are recovered by the recall procedure from detected ultrasonic signals and vice versa. Furthermore, from the discrepancy between the presented and the learned signals, the changes in the wave phenomenon, corresponding, for example, to changes in the boundary conditions of a specimen, can be determined.

Experimental characterization of ultrasonic phenomena by a neural-like learning system

“…Another application of the processing scheme has been to characterize the components of a vector source. For this, the fracture of pencil leads activated in various directions along one edge of a 3-inch thick aluminum block was used [6]. An array of four sensors on the top surface of the block detected the simulated AE.…”

“…1. Its features, which have been described in several publications [6,7,8], use some ofthe fundamental principles of neural networks [9]. Briefiy, we assurne that an ultrasonic wavefield can be characterized by a finite set of data supplied from an array of sensors together with selected features of the source, structure or sensor.…”