Damage identification of adhesive bonded structures based on ultrasonic guided waves and neural network

Adhesive bonded structures are widely used in aerospace and other industries due to their lightweight and high-strength properties. However, interfacial debonding remains a critical failure mode that compromises structural integrity. Although ultrasonic guided waves can detect such damage, existing methods lack robust quantitative analysis capabilities. To address this limitation, we propose a deep neural network – based framework for precise debonding identification using guided wave signals. A validated finite element model generates training data covering various debonding scenarios, enabling the system to map wave signals to structural images. The results indicate that this method can accurately identify debonding damage, achieving an accuracy rate of 0.9977 on the test set. Upon verification, it has been confirmed that the maximum localisation error for debonding damage does not exceed 3.16 mm. Additionally, concerning size recognition, the Dice coefficient for circular debonding damage is found to be no less than 0.8950, while for rectangular debonding damage, it stands at no less than 0.7820. Tests on three adhesive structures confirm its diagnostic reliability, demonstrating significant progress in debonding identification for structural health monitoring.