Acoustic emission classification using signal subspace projections

In using acoustic emissions (AE) for mechanical diagnostics, one major problem is the differentiation of events due to crack growth in a component from noise of various origins. This work presents two algorithms for automatic clustering and separation of AE events based on multiple features extracted from experimental data. The first algorithm consists of two steps. In the first step, the noise is separated from the events of interest and subsequently removed using a combination of covariance analysis, principal component analysis (PCA), and differential time delay estimates. The second step processes the remaining data using a self-organizing map (SOM), which outputs the noise and AE signals into separate neurons. The algorithm is verified with two sets of data, and a correct classification ratio of over 95% is achieved. The second algorithm characterizes the AE signal subspace based on the principal eigenvectors of the covariance matrix of an ensemble of the AE signals. The latter algorithm has a correct classification ratio over 90%.