Physics-based classification of acoustic emission waveforms

Abstract

The classification of acoustic emission source mechanisms based on features related to the physics of acoustic emission signal generation is considered in this paper. Numerically generated acoustic emission waveforms are used for this purpose. Conventional acoustic emission parameters such as rise-time, duration, and frequency content do not effectively characterize acoustic emission waveforms for the purpose of identifying the source mechanisms. Features unique to the different source mechanisms and relative positions of the sensor with respect to the source were identified and extracted from numerically obtained acoustic emission waveforms. This feature selection appears to be successful in capturing the differences related to the source mechanisms considered here. Correlation coefficients of the 45 features with different waveforms were first obtained, and their principal components determined. The dominant principle components were found to adequately characterize the waveforms and relate them to their source mechanisms. Better than 90 percent success was seen when only the first two principle components were employed, even in noisy signals considered here.