Research of Acoustic Emission Characteristics and Applicability of Artificial Intelligence for Source Localization

Within this research, the deployment of artificial intelligence for the localization of acoustic emissions (AE) is being investigated. In detail, series of experiments were conducted on a plate of granite with homogeneous and isotropic material behavior. Acoustic emissions could be generated by the breaking of pencil leads, as well as the impact of rice grains and steel balls on 81 specific positions with a distance of 50 mm in both horizontal axes to each other. The response of the system was recorded using 16 acoustic emission sensors, of which 8 sensors were positioned on each surface side of the plate. In further steps the data could be processed, so that the resulting characteristic signal features could be translated into spec-trograms and furthermore being used as input for the artificial intelligence. Additionally, the corresponding source locations appeared as labels. Both, the signal features as well as the labels, could be implemented in the process of supervised learning using a convolutional neural network (CNN). The results of this network indicated a validation accuracy of over 99% as well as a low loss value by classifying the inserted features to the right labels. Thus demonstrates the excellent applicability of artificial intelligence for the source localization of acoustic emission (AE).