Multievent localization for loop-based Sagnac sensing system using machine learning.

Area of Science:

• Optical sensing
• Fiber optic sensors
• Machine learning

Background:

• Accurate event localization is critical for pipeline monitoring and intrusion detection.
• Existing methods face challenges in multievent localization over long distances.

Purpose of the Study:

• To experimentally demonstrate multievent localization for long perimeter monitoring.
• To apply machine learning techniques, specifically deep neural networks (DNNs), for precise event location.
• To address the challenge of distinguishing and locating multiple simultaneous events in an optical fiber sensor.

Main Methods:

• Utilized a Sagnac interferometer loop sensor with over 100 km of single-mode fiber.
• Treated multievent localization as a multilabel multiclassification problem across 250 fiber segments.
• Developed a deep neural network (DNN) model trained on simulated data, with complexity reduced using discrete cosine transform (DCT).

Main Results:

• Achieved 99% accuracy for single-event localization within one segment error in simulations.
• Demonstrated 95% accuracy for localizing one of two events and 78% accuracy for localizing both events within one segment error.
• Experimental results validated simulation findings, confirming the model's effectiveness for high-precision event localization.

Conclusions:

• The proposed DNN-based approach effectively enables accurate multievent localization in long-perimeter optical sensing.
• The method shows significant promise for applications like pipeline monitoring and intrusion detection.
• Future work can extend the model to detect and localize more than two simultaneous events.