ResDAU-Net AE signal pickup

Abstract

In order to improve the pickup efficiency of AE signals, the sampling part was replaced with residual units based on the U-Net network structure, and the PAM and CAM dual attention modules were embedded in the jump structure to form the ResDA-UNet network. Input the AE signal waveform data, output the probability distribution of P wave, S wave and noise, and pick it up at the maximum probability point. By comparing the AR-AIC and reference threshold STA / LTA picking method, the proportion of the absolute errors reached 5ms, 15ms, 25ms and 50ms, and the proportion of ResDAU-Net network reached 93.5%. Compared with traditional ResDAU-Net methods, the picking accuracy and picking speed are higher, and the network has higher adaptability to feature extraction of acoustic emission signal waveform data.