Acoustic emission source localization by artificial neural networks

Abstract

The objective of this work is to present an alternative localization method based on the use of neural networks, using experimental training data as a modeling basis. For this purpose, test sources are applied on the test object to yield input data for a neural network. Subsequently, the trained neural network can be applied to recorded data from material failure of the test object. The presented method is validated using a type III carbon-fiber-reinforced polymer pressure vessel with metallic liner and is compared with an established localization method using the time difference of arrivals. It was shown that the neural-network-based method is not only superior by a factor of 6 in accuracy but also results in a lower scattering of the localized source positions by a factor of 11. For the neural-network-based approach, the localization accuracy is only limited by the theoretical localization accuracy, which is based on measurement errors of the acquisition chain and the subsequent determination of the time of arrival of the detected signal. Source localization using neural networks on the basis of experimental training data thus is very promising to approach the limits of theoretical measurement accuracy.