Early Warning of Anomalous Events Using Optical Fiber

Rockfalls and landslides result in severe danger to infrastructure and public safety in mountainous regions. Detecting these events in real time is crucial for risk mitigation and early warning systems. This thesis investigates the use of state of polarization (SOP) data taken from an experimental site situated in a mountain gully, to monitor and detect anomalous events, particularly falling rocks. The study is based on data collected from four distinct fiber optic circuits, either buried or exposed, deployed along the gully. These circuits continuously record SOP angular speed (SOPAS) values, which represent the dynamic variations in the polarization state of light traveling through the fibers. The raw data are continuously acquired by polarimeter, which then processed and analyzed using Matlab software. The main aim of this research study is to build an efficient data processing system to analyze SOPAS data, identify anomalies, and distinguish natural environmental variations from significant rockfall events. The methodology includes signal filtering techniques such as Savitzky-Golay and moving average filters to reduce noise and remove system bugs while preserving key event characteristics. A combination of statistical analysis—such as mean, standard deviation, and percentile calculations—and threshold-based anomaly detection is applied to characterize event patterns. Furthermore, the study explores the inverse cumulative distribution function (ICDF) of SOPAS values to examine the tail of distribution to emphasize rare but high-intensity fluctuations. To validate the proposed detection methodology, two distinct datasets were analyzed: (1) a quiet period with minimal disturbances, serving as a baseline, and (2) an experimental rockfall scenario, where instrumented rocks were deliberately dropped onto the fiber circuits to generate controlled SOPAS. The results highlight clear differences between natural background noise and induced anomalies. Scatter plots of anomalies durations versus threshold levels provide further insights into the longevity and intensity of detected anomalies. The findings demonstrate that SOPAS analysis offers a promising approach for rockfall detection in fiber optic sensing applications. The developed framework effectively identifies rockfall events, paving the way for the integration of polarization-based sensing technology into early warning systems for rockfall-prone regions. Future work may focus on refining detection algorithms, integrating machine learning techniques, and expanding the deployment of fiber optic circuits to broader geographic areas for enhanced monitoring capabilities.