Cyclic response of coral concrete: Coupled mechanical-acoustic emission analysis and machine learning-enabled damage identification

This study investigates the mechanical properties of coral concrete with different sizes under uniaxial cyclic loading, including peak stress, peak strain, stiffness degradation, plastic strain, stress degradation, and energy dissipation. Acoustic emission (AE) parameters, such as ringing counts and Felicity ratio, are analyzed to establish the relationship between these parameters and the damage state. The results show that with the increase in cyclic loading, the strength and stiffness of coral concrete degrade progressively, while plastic strain and stress attenuation significantly affect its long-term service performance. Given the limitations of traditional AE-based damage evaluation methods—such as relying on single parameters like b-value or cumulative ringing counts, which provide limited information, are sensitive to noise and testing conditions, and fail to capture the complex temporal patterns of damage evolution—this study adopts a machine learning approach with strong multi-feature fusion and temporal modeling capabilities. By integrating multiple AE parameters with a Bi-LSTM neural network, a damage identification framework for coral concrete under cyclic loading is established, enabling accurate modeling and prediction of damage evolution. The findings offer practical technical support for condition assessment and durability management of coastal engineering structures.