Lamb Wave Dispersion Compensation Based on a Fourier Basis Convolutional Autoencoder

Abstract

The inherent dispersion of Lamb waves will reduce the signal-to-noise ratio (SNR) and detection sensitivity of the detection signal, seriously affecting the resolution of defect identification. Therefore, it is necessary to design an effective method to reduce the influence of the dispersion effect. Traditional dispersion compensation methods are generally restricted to single-mode Lamb waves and depend heavily on manual extraction of signal characteristics; this limitation greatly reduces the generalization ability of the dispersion compensation model. In recent years, deep learning has attracted widespread attention in various fields due to its excellent adaptive feature extraction capabilities. Therefore, this article proposes a Lamb wave dispersion compensation based on a Fourier basis convolutional autoencoder (FCAE) network. Considering the frequency correlation of Lamb waves, the Fourier basis is introduced into the convolutional autoencoder (CAE) so that the network model can combine the time-frequency domain characteristics of the signal, learn the flight time of the wave packet of the dispersive signal, and reconstruct the nondispersive Lamb wave in combination with the excitation signal’s waveform. Compared with traditional dispersion compensation methods, this method can achieve dispersion compensation of Lamb waves in multimode and multiwave packet situations. Through numerical simulation and experimental verification, and comparison with various deep learning models such as convolutional neural networks (CNNs) and hole CNNs, it was verified that the proposed model still has good performance under different numbers of wave packets.