Delamination Damage Detection of CFRP Composite Structures Using DSAN-based Deep Transfer Learning Approach

Abstract

Carbon fiber reinforced plastics (CFRP) is a kind of lightweight composite material widely used in aerospace. Since the progressive development of fatigue damage is complex and leads to potential safety risks in CFRP structures, structural health monitoring based on Lamb wave has been developed to track the growth of fatigue damages using a sensor network attached to the surface, which is experimentally intensive and expensive. To overcome the above challenges, a composite fatigue damage diagnosis method based on deep transfer learning is proposed to transfer the physical mechanism provided by numerical models to the diagnosis of real monitoring data. Firstly, numerical models of the composite structures are built to indicate the accumulation of fatigue damage during the full life cycle by introducing delamination between three sub-structures of cross-ply laminates. Then simulation signals with high fidelity are generated by virtual sensors and input into a data-driven diagnostic model with monitoring data. By aligning the data distribution of corresponding categories in simulation and experiment datasets respectively, the sub-domain adaptation is implemented and the physical mechanism provided by digital models is thereby fused with monitoring data. In this situation, the diagnostic model can still achieve more than 81% accuracy on a smaller training set, which performs better than conventional methods and significantly reduces the number of aging experiments.