Real-Time Vibration Anomaly Detection in Operational Metropolitan Optical Networks Using Machine Learning

This paper presents an experimentally validated machine-learning-based framework for real-time detection and classification of vibration-induced anomalies in a live metropolitan optical transport network. The proposed approach exploits high-frequency state-of-polarization (SOP) fluctuations as an early-warning metric to distinguish critical, potentially harmful vibrations from benign environmental noise. The framework is experimentally validated on a 28.5 km deployed metro fiber link operating under real traffic conditions, where vibrations are intentionally induced to mimic realistic threats such as drilling near buried fibers. A supervised machine-learning pipeline is applied to analyze SOP dynamics and classify vibration events under noisy operational conditions. Multiple classifiers, including Decision Tree, Random Forest, and XGBoost, are evaluated, with XGBoost achieving the best performance, reaching an accuracy of 98.02% and an AUC of 0.987. The results demonstrate reliable discrimination between normal and critical vibration events while maintaining low computational complexity, enabling practical real-time edge deployment. The proposed framework supports proactive fault prevention and enhances the resilience of metropolitan optical networks.