Unsupervised Damage Localization In Composite Plates Using Lamb Waves And Conditional Generative Adversarial Networks

Composite plates are increasingly used in several engineering fields. A common way for monitoring the health state of these structures is by analysing ultrasonic guided waves propagating in the plate. Among guided waves, Lamb waves (LWs) have shown promising diagnostic capabilities, and have been recently used for damage diagnosis in deep learning-based frameworks. However, so far, the proposed frameworks have mainly leveraged supervised algorithms, which require acquiring and labelling a large amount of data when the structure is in healthy and damaged conditions. Besides requiring much time and effort, acquiring enough data in damaged structures may not be practical in real world. Hence, this paper proposes the use of conditional generative adversarial networks (CGANs) with convolutional layers for damage localization in composite plates. As unsupervised algorithms, CGANSs can be trained on LWs acquired when the structure is healthy, and do not require information about damaged states. The proposed method was validated through an experimental case study involving two different composite plates.