A neural network-based air-coupled ultrasonic damage detection method for composite honeycomb sandwich structure

Fiber reinforced composites have been widely used due to their excellent mechanical properties and designability, but the damage forms of composite materials are more complex compared to metals. Regular non-destructive testing (NDT) is crucial to ensure the reliability of equipment during use. Air-coupled ultrasound testing (ACUT), which uses air as the coupling agent, has the advantages of non-contact, simple equipment, and no pollution, and is widely used in NDT of composite material structures. However, the difference in acoustic impedance between the specimen and air results in poor accuracy of ACUT detection, and the analysis of the results is influenced by the experience of the technical personnel. To address this issue, a neural network-based method for damage detection and imaging has been proposed. Firstly, ACUT was used to detect pre-embedded defects at different depths within the composite honeycomb sandwich structure, and the original ultrasonic signals were obtained. Then, based on the pattern of sudden changes in signal amplitude at the defect boundary, detailed signal classification and defect area division were carried out by superimposing the results of damage area judgment under different thresholds. Finally, the signals after augmentation were input into neural networks to identify the state of the structure, and the prediction results are output to a pixel matrix to form a detection image. The results demonstrate that the trained neural network has a high classification accuracy; Even when faced with unseen data, the network still has strong robustness and generalization. Compared with traditional methods, neural networks only require A-scan signals to achieve state prediction, without the need for signal comparison in different states and relying on human experience, which can effectively improve recognition efficiency and reliability.