A rapid multi-defect imaging method for composite plates using air-coupled ultrasonic guided waves based on...

Carbon fiber composite materials are widely used in the aviation industry as a key technology for aircraft lightweight. As structures evolve towards larger, more complex, and intelligent designs, there is a growing need for rapid and accurate detection of multiple and micro-scale defects. However, traditional damage index methods, which rely on manual feature extraction, often encounter problems such as numerous artifacts and blurred boundaries when identifying multiple defects, thereby limiting detection accuracy and automation levels. This paper proposes a rapid multi-defect imaging detection method for composite plates using air-coupled ultrasonic guided waves, based on a multi-scale interactive Siamese network. First, the propagation characteristics of guided waves are analyzed through numerical simulation to determine the optimal probe incident angle and guided wave mode. Second, a multi-scale feature interaction and fusion module is constructed to automatically extract feature vectors between baseline (healthy) samples and test samples. Subsequently, a Siamese network framework incorporating this multi-scale feature interaction and fusion module is developed. The network is trained using a cross-entropy loss function to distinguish between normal and abnormal states, thereby enhancing the Euclidean distance between features of damaged samples and internal healthy samples. This Euclidean distance serves as the damage index. Finally, probability damage curves are calculated in the 0°, 45°, 90°, and 135° directions and extended into an imaging matrix. The matrices are divided into two groups based on their orthogonality. The damaged image is obtained by first summing the matrices within each group and then multiplying the resultant matrices between groups. Experimental results show that the proposed method effectively enhances the distinction between healthy and damaged areas while suppressing artifact generation in healthy regions. The maximum central positioning errors for single simulated damage and delamination damage are 1.8 mm and 5.5 mm, respectively, and the maximum central positioning error for multi-defect detection is 3 mm. Compared with point-by-point C scanning, the method proposed in this paper increases the scanning efficiency by approximately 168.9% per unit time, meeting the requirements for rapid screening and preliminary diagnosis of damage in large composite structures.