A Lightweight Network for Defect Detection in Polyethylene Pipelines Using Multi-Mode Ultrasonic Guided Waves

Ultrasonic Guided Wave (UGW) has been widely applied in the field of nondestructive testing (NDT) for polyethylene (PE) pipelines due to its long-range inspection capability and high sensitivity to defects. In recent years, the integration of deep learning techniques has further advanced UGW-based defect detection by enabling more accurate and automated signal interpretation. However, most existing deep learning methods predominantly focus on single-mode UGW signals and involve high computational complexity, which significantly limits their effectiveness in complex and resource-constrained real-world scenarios. To address these challenges, this paper proposes a lightweight deep learning network, termed UMENet, specifically designed for multi-mode UGW-based defect detection. It integrates a mode-aware feature enhancement block and a Multi-Scale Depthwise Separable Convolution (MDS) branch to emphasize discriminative characteristics and extract multi-scale features from different UGW modes. This design balances local and global information fusion, achieving high accuracy and lightweight architecture for PE pipeline inspection. Experimental results demonstrate that the proposed network achieves superior performance across multiple evaluation metrics, with only 331K parameters, an accuracy of 98.75%, a precision of 98.52%, a recall of 98.70%, and an F1-score of 98.59%. Furthermore, this study conducts extensive comparative experiments involving multiple UGW modes and several state-of-the-art deep learning models to comprehensively validate the superior performance of UMENet under various signal conditions. The results also identify the most suitable guided wave mode for defect detection in PE pipelines, providing strong guidance for practical mode selection in UGW-based applications.