METHODS FOR REMAINING USEFUL LIFE PROGNOSIS OF ADHESIVELY BONDED COMPOSITE STRUCTURES

This paper introduces a novel approach for assessing damage and predicting the remaining useful life (RUL) of adhesively bonded fiber-reinforced polymer (FRP) structures. It combines fracture mechanics (FM) and deep learning (DL) techniques grounded to emulate the fatigue and damage behaviors of the adhesively bonded composite material. Furthermore, the paper presents a crack growth simulation methodology that merges the extended finite element method (XFEM) and fracture mechanics to theoretically predict damage extent. It redefines the relationship between bonded area degradation and fatigue cycles by considering crack propagation direction. Lastly, the paper leverages a database of hybrid model samples to integrate Dropout and Monte Carlo methods, replacing traditional Bayesian dynamic network quantification of uncertainty. This enhancement allows the DL network to generate the RUL along with an associated confidence interval for uncertainty. The results underscore the advantages of this hybrid model in accurately predicting RUL and identifying damage in FRP structures.