A Method for Arrival Time Picking of Acoustic Emission Signals Based on Convolutional Neural Networks

Abstract

To achieve an accurate pickup of acoustic emission (AE) signal arrival time, a convolutional neural network (CNN)-based AE arrival time pickup method is proposed. The positioning results of the concrete lead fracture test were used to verify the arrival time picking accuracy of the CNN algorithm, and it was compared with the threshold-based arrival time extraction algorithm. The results showed that after applying the CNN algorithm for correction, the number of AE events increased. The distribution of positioning points after arrival time correction more accurately reflects the actual pattern, with clearer spacing between vertical lines and points clustered near grid points. From the error distribution map, the average positioning error using threshold-based arrival time data is relatively large, with more yellow points on the map and usually lighter colors. The CNN model significantly reduced the average error of lead fracture points, and the corrected average positioning error at the time was reduced by about 6.8 mm, achieving excellent results. These findings can provide reference for the selection of AE signal arrival time.