Research on Pipeline Damage Identification Based on Optical Fiber Sensing and Deep Learning

Pipelines are prone to safety incidents due to accumulated damage and defects during long-term operation. Although distributed fiber optic sensing technology has received considerable attention for leakage monitoring, research on damage detection in the early stages of leakage remains insufficient. Therefore, this paper proposes a pipeline damage identification model based on CNN-BiLSTM-SKNet multi-feature fusion. The model utilizes vibration signals collected by a distributed vibration fiber optic sensing system under flow-induced vibrations as input. An improved SKNet module is incorporated to introduce an attention mechanism, enabling adaptive weight allocation to feature maps extracted by multi-channel convolution. Additionally, skip connections are introduced to preserve relevant information. Experimental results show that the proposed model achieves an average recognition accuracy of 93.65% under four operating conditions. This study not only provides a novel technical solution for pipeline health monitoring based on distributed fiber optic sensing but also demonstrates promising application potential.