Development of an artificial neural network for source localization using a fiber optic acoustic emission sensor array

Abstract

An intelligent algorithm was developed based on backpropagation artificial neural network for the acoustic emission source localization. The established and trained methods of the algorithm were stated with the time difference of arrival detected by a fiber optic acoustic emission sensor array and the coordinates of acoustic emission source. The response characteristic of fiber optic acoustic emission sensor was calibrated with the commercial piezoelectric ceramic transducer (PZT) acoustic emission sensor, which provided that the fiber optic acoustic emission sensor was better suited to detect the low frequency of stress wave than the PZT sensor. Four fiber optic acoustic emission sensors were deployed in a square array in an aluminum plate for comparisons between different algorithms of source localization. Comparison results of acoustic emission source location provided that the intelligent algorithm improved the accuracy by reducing the nonlinear errors. For the anisotropic materials, a sensor array deployed in a diamond pattern was adopted. The velocities of stress waves in orthogonal directions were measured as the basic performance for both algorithms of source localization. Four sensors were integrated into a carbon fiber–reinforced polymer plate as a perfect structure for locating the acoustic emission source impacting on its surface. The experiment results provide that the maximum error is only 6.3 mm using the intelligent algorithm.