Distributed acoustic sensing system for linear asset monitoring: Signal modeling, threat detection and classification

The fiber-optic distributed acoustic sensing system based on the Phase-OTDR technique provides a cost-effective solution for monitoring long linear assets such as oil and gas pipelines, power lines, railway lines, boundaries of medium to large-scale facilities. Fiber optic distributed acoustic detection (DAS) can use telecommunications class fiber optic cables as detection and transmission media. Extinction ratio is an inherent limiting factor that has a direct effect on the detection performance of phase-OTDR based distributed acoustics sensing systems. In the first part of the thesis, a model based analysis of Rayleigh scattering is presented to simulate the effects of extinction ratio on the received signal under varying signal acquisition scenarios and system parameters. These signal acquisition scenarios are constructed to represent typically observed cases such as multiple vibration sources cluttered around the target vibration source to be detected, continuous wave light sources with center frequency drift, varying fiber optic cable lengths and varying ADC bit resolutions. Results show that an insufficient extinction ratio can result in high optical noise floor and effectively hide the effects of elaborate system improvement efforts. The second part of the thesis adopts two main approaches: an adaptive power thresholding approach for real time threat detection on fiber optic based distributed acoustic sensing systems, and a real time threat classification approach to be used in a distributed acoustic sensing system that is developed for monitoring linear assets with a maximum length of 50 kms is explained. Due to the lack of an approved statistical signal model, and the changes on measured signal magnitude, which result from the internal mechanism of the optical system and physical activities that occur near the fiber optic cable, signal statistics are directly generated from the captured data. Since it is not known whether the captured data contain only noise or noisy activity data, power threshold is computed by applying two step statistical model approach. In order to construct these statistical models, first and second order Gaussian mixture models are tested and the most appropriate is used for computing power threshold. Finally, the update mechanism of power thresholding is explained and the execution time of the algorithm is analyzed. A deep learning approach based on The Convolutional Neural Network (CNN) is adopted for threat classification. Accuracy of the classification and durations of execution for neural networks with varying architectures and complexity are computed. A proposed approach for classifying all the detected threats without decreasing the detection accuracy is introduced. The maximum allowable execution time for the network structure that is appropriate for the proposed approach is analyzed for the worst case scenario. Hence, the most appropriate network architecture selection can be performed based on classification accuracy and also applicability in real-time criterion.