Monitoring of abnormal conditions of underground pipelines using fiber-optic vibration sensing and deep learning

Abstract

Pipelines are widely used for the long-distance transportation of fossil fuels and industrial materials. For underground pipelines, the pipeline safety is seriously threatened by various abnormal conditions, including pipe leakage, mechanical impact, earth excavation, third-party intrusion and natural disaster. Therefore, real-time monitoring of abnormal conditions is an effective way to ensure the safety of underground pipelines. In this paper, a fiber-optic vibration sensing system is used to monitor underground pipelines. Deep learning-based methods are proposed for detection and recognition of abnormal pipeline conditions. The detection method integrates a deep scattering network (DSN) with kernel support matrix domain description (KSMDD). The DSN firstly extracts the deep time-frequency features of vibration signals obtained by the fiber-optic sensing system. The KSMDD is then applied on signal feature matrices to define a statistical index for abnormal condition detection. The recognition method integrates a DSN with a gated recurrent unit (GRU) network to extract signal features for recognizing different types of abnormal conditions. Experimental results illustrate that the abnormal condition detection and recognition methods have a false alarm rate of 1 %, a detection rate larger than 99 %, and a recognition accuracy higher than 99 %.