Guided Waves-Based Disbond Detection of Double-Layer Plates using LSTM Networks

Abstract Adhesively bonded structures are of great interest in a wide range of industries. However, such adhesive bonds are prone to interfacial defects like disbond and delamination during both the fabrication process and service life. In this paper, we propose a deep-learning (DL) approach to automatically localize and size the disbond in a double-layer plate using ultrasonic guided waves. This plate consists of an aluminum substrate with a stainless-steel coating layer. A guided wave active sensing procedure is used by implementing one transmitter-receiver configuration. Both guided wave pulse-echo and pitch-catch signals are simulated through finite element simulations under various disbond scenarios. To account for uncertainty and noise in the experimental measurements, Gaussian random noise is introduced in the numerically simulated data. The proposed DL model organically combines the convolutional neural network (CNN) with long short-term memory (LSTM). Once trained, the neural network is capable of outputting the location and length of the disbond between the transmitter and receiver. Not only the testing set but also the extended unseen dataset is accurately predicted by the well-trained neural network. These results demonstrate that the proposed method has tremendous potential for characterizing disbond in practical nondestructive evaluation (NDE) and structural health monitoring (SHM) applications.