Deep learning-enhanced acoustic emission localization for damage evolution analysis in coal-concrete composite structures under cyclic loading

Renbo Gao, Heping Xie, Fei Wu, Cunbao Li, Chunfeng Ye, Yue Liu,
Deep learning-enhanced acoustic emission localization for damage evolution analysis in coal-concrete composite structures under cyclic loading,
Engineering Structures,
Volume 343, Part C,
2025,
121208,
ISSN 0141-0296,
https://doi.org/10.1016/j.engstruct.2025.121208.
(https://www.sciencedirect.com/science/article/pii/S0141029625015998)
Abstract: The stability of coal-concrete composite structures plays a vital role in the reuse of underground coal mine spaces. To elucidate their damage evolution under cyclic loading with increasing upper limits, this study integrates laboratory testing, computed tomography scanning, and acoustic emission monitoring to systematically investigate the mechanical behavior and failure modes of specimens with varying height ratios. To address the limited accuracy of traditional acoustic emission source localization methods, a convolutional neural network model integrating a squeeze excitation module and multi-head self-attention mechanism is proposed. Comparative results demonstrate that this model outperforms conventional deep learning approaches in both localization accuracy and convergence efficiency. Environmental disturbance experiments further validate the model’s adaptability, showing that drying and water immersion treatments significantly alter acoustic emission signal characteristics and amplify the material’s nonlinear response. At the microscale, scanning electron microscopy imaging combined with acoustic emission localization reveals that stress concentration at the coal-concrete interface triggers wedge-shaped failure, which acts as the dominant mechanism of structural failure. Specimens with different height ratios exhibit distinct failure modes. This study offers a novel approach for damage identification in composite structures within underground coal mine spaces and provides a technical foundation for intelligent monitoring and predictive maintenance.
Keywords: Coal-concrete structure; Acoustic emission localization; Deep learning; Failure mode