Third party interference is one of the leading causes of pipeline failures and accidents that create great risk for safety and environment, as well as revenue loss for the operators. Especially in urban areas unauthorized and uncoordinated infrastructure and construction works pose serious threat to liquid and gas pipelines. Effective detection of intrusive activities and a timely preventative course of action is crucial. Fiber Optic Distributed Acoustic Sensors (DAS) are useful and proven tools for third party interference detection.
Distributed Acoustic Sensing (DAS) is a sensing technology that uses standard telecommunications fiber optic cable that is buried parallel to the pipeline as an array of acoustic sensors. Detection and classification of intrusive activities in urban areas while minimizing nuisance alarm rates is a challenging problem due to high intensity urban activity noise. A robust solution in the urban environment requires multi stage machine learning (ML) approach that takes advantage of state-of-the-art sound activity recognition and anomaly detection algorithms.
We propose a three-stage interference detection algorithm. The first stage of the algorithm is sound recognition that classifies short sound clips with Convolutional Neural Networks (CNN). The second stage of the algorithm is model based event detection that recognizes activity patterns in series of audio clips. The final stage of the algorithm fuses data from multiple channels of the acoustic array to model moving vibration sources to reduce the nuisance alarms.
Proposed algorithm is deployed in highly populated urban areas for securing natural gas pipeline in Istanbul, Turkey. The system has successfully detected multiple instances of third-party interferences with a very low nuisance alarm rate.