"Damage Detection in Small-Diameter Pipes Using Ultrasonic Guided Waves "

"This study uses deep learning techniques and ultrasonic guided waves (UGW) to investigate the detection of damage in small-diameter pipes. UGW propagation in small-diameter pipes is complex  and noise-susceptible, leading to inaccurate damage detection. Therefore, constructing an effective deep learning model for damage detection is crucial. We propose a multi-channel attention Mamba-2 (MCA-Mamba-2) network for precise identification of pipe damage severity. This model performs multi-channel feature extraction on the time-series waveforms of ultrasonic guided waves propagating through the pipe, assigns weight factors according to channel importance, and achieves end-to-end damage identification. A dataset comprising ten distinct damage categories is constructed. In the dataset, the proposed model achieves detection accuracies of 81.09\\%, 83.65\\%, 82.49\\%, and 84.15\\% on 0 dB noisy test sets with white noise, pink noise, Gaussian noise, and Laplacian noise, respectively, outperforming the baseline Mamba-2 model by up to 13.36\\%. The model effectively learns the complex mapping between ultrasonic guided waves and pipe damage, thereby demonstrating exceptional robustness in experiments and enabling accurate damage identification under various noise conditions."