Data-driven generalization for guided wave damage identification across beam-like structures

Data-driven approaches have been applied to structural health assessment due to their capability to address complex challenges and achieve high-precision and rapid defect detection. This article presents a deep learning-based generalization framework for guided wave (GW)-based damage identification in diverse isotropic beam-like structures with various cross-sectional geometries and material properties. A multiscale convolution and squeeze-and-excitation attention-assisted bidirectional hybrid (MC-SE-BiH) network is proposed to generalize the damage identification. A higher-order beam model is employed to enhance computational efficiency and simulation accuracy of modelling GW propagation in various damage scenarios. A series of numerical case studies is carried out to demonstrate the generalization ability and robustness of the proposed MC-SE-BiH framework in damage identification using GW. The effectiveness and practicality are further validated using experimental measurements obtained from isotropic beams with different material properties and cross-sectional configurations. The results demonstrate that the proposed MC-SE-BiH framework can identify damage across diverse beam-like structures.