Fatigue evolution prediction for fiber‐reinforced plastics based on frequency‐wavenumber wavefield of guided wave using deep‐learning model

Abstract

Employing the laser ultrasonic system, information on fatigue evolution can be captured and explicitly stored in the frequency‐wavenumber wavefield of the guided wave. This paper presents a deep‐learning architecture for processing the frequency‐wavenumber wavefield, which comprises two distinct steps: fatigue characterization and evolution prediction. Firstly, the fatigue characterization step employs a convolutional autoencoder (CAE) for compressing the frequency‐wavenumber wavefield and a fully connected network (FCN) for obtaining the fusion fatigue characteristic. Due to the high cost of experimental samples, extensive simulation‐generated wavefields are used to pre‐train the network. Subsequently, the fatigue evolution prediction model based on the latent ordinary differential equation (Latent‐ODE) is trained for the step‐by‐step prediction of fatigue evolution with a small amount of composite fatigue test data. The results validate the effectiveness of the deep‐learning architecture in characterizing fatigue and predicting its evolution, as well as the feasibility of the frequency‐wavenumber wavefield compression of guided wave.