Damage localization method using ultrasonic lamb waves and Wav2Vec2.0 neural network

In this paper, a Wav2Vec2.0 neural network based on an attention mechanism is proposed to locate defects in array ultrasonic testing signals. This method does not require knowledge of the priori condition sample sound velocity or feature extraction scattering First, piezoelectric system used detect signal through hole at different positions plate structure. Then, three networks—1D-CNN, Muti-Transformer, and Wav2Vec2.0—are collected data. The performance verified with data set finite element simulation experimental system, identification accuracy calculation efficiency networks are compared analyzed. To provide solution for poor balance weak noise resistance set, expansion time domain transformation technology proposed. research results show that, positioning article 98.46%, superior Muti Transformer ID-CNN.