Experimental characterization of ultrasonic phenomena by a learning system
Igor Grabec, Wolfgang H. Sachse · Journal of Applied Physics · 1989
This paper describes the application of an adaptive learning system comprising an associative memory to the characterization of ultrasonic phenomena. The mapping of source and waveform data and vice versa can be performed by utilizing the off-diagonal cross-correlation portions of the associative memory. It is suggested that a more general description of the ultrasonic phenomena can be obtained if the diagonal, autocorrelation portions of the memory are also utilized. In this case, the memory is applicable for the autoassociative optimal filtering of experimental data. Experiments are described which utilize such an adaptive system, running on a laboratory minicomputer, to process the signals from a transient ultrasonic source in a plate specimen. It is shown how the system learns from the experimental pattern vectors, formed from the ultrasonic waveforms and encoded information about the source. The source characteristics are recovered by the recall procedure from the detected ultrasonic signals and vice versa. Also, the changes in the wave phenomenon corresponding to changes in the boundary conditions of the specimen can be detected from the discrepancy between the presented and the learned signals.