Bidirectionally transformed subspace cross-domain sparse representation for CFRP fatigue damage identification under different operating conditions

Lamb waves have been established as a reliable choice for identifying fatigue damage in carbon fiber-reinforced polymer (CFRP). In practice, Lamb wave signals are collected under both non-fatigue and fatigue loading conditions, which significantly affects the propagation of Lamb wave. Furthermore, the signal variations caused by the above two operating conditions resemble those induced by fatigue damage. These changes are mainly reflected in amplitude variations and phase shift, which complicates the accurate identification of fatigue damage states under varying loads. This paper aims to eliminate the interference of loading conditions through domain adaptation, while simultaneously identifying the fatigue damage states using sparse representation. We presented and verified an integrative bidirectionally transformed subspace cross-domain sparse representation method. In order to enhance interference elimination, signals from different loading conditions are bidirectionally transformed into a common subspace. This transformation allows for a broader adjustment range, and further minimizes the domain discrepancy. To improve the damage identification performance, we extract signal features within the subspace using sparse representation and incorporate a linear classification module. The variables for domain adaptation, sparse representation, and linear classification module, are solved in two distinct optimization steps. The robust relationship between domain adaptation and classification enhances the overall damage identification performance. The proposed method is formulated as a constrained optimization problem, and the corresponding solution strategy is precisely derived. In order to validate the proposed method, extensive experiments were conducted using NASA-published CFRP dataset. The results demonstrate that the proposed method effectively eliminates the interference of loading conditions, achieving an average damage identification accuracy of 86.57%. This outperforms other state-of-the-art models and demonstrates excellent robustness.