Meta-Transfer-Learning based Damage Detection of CFRP Composite Structures

Abstract

Applying data-driven approaches in damage detection of CFRP composites is becoming increasingly popular with the rapid development of deep learning methods. However, obtaining enough data for training these data-driven models is challenging, and the presence of imbalanced data can further exacerbate the problem. Moreover, due to the limited availability of CFRP data under various structural conditions, it is desirable to make the most use of the existing data or leverage the previously learned models. To handle these problems, we propose a transfer learning-based approach, which combines the benefits of transfer learning to overcome the challenges caused by limited data, and benefits of meta training to effectively train new models. Our experiments demonstrate the efficacy of this approach in identifying damage in CFRP.