Clustering Analysis of Acoustic Emission Signals Based on Unsupervised Deep Learning

In order to improve the automation degree of detecting and determining the damage types of metal structures by acoustic emission signals, starting from the acoustic emission characteristic parameters, this paper proposes an acoustic emission signal recognition method based on deep-learning neural networks and cluster analysis. Aiming at the problem that it is difficult to obtain data labels for acoustic emission signals, an unsupervised learning method is adopted. Deep-learning training is carried out according to a large number of actually measured acoustic emission parameters, and then an attempt is made to combine the K-means clustering algorithm to distinguish different types of acoustic emission signals. An experiment was designed to perform cluster analysis on the acoustic emission signals of lead-break signals, steel-ball-drop signals, knocking signals, and metal-friction signals on a metal plate. The results show that the proposed method can automatically recognize different categories of acoustic emission signals, and the recognition effect is better than that of the clustering method based on artificially set acoustic emission signal characteristics.