Hybrid CNN-Ensemble Framework for Intelligent Optical Fiber Fault Detection and Diagnosis

Ensuring the integrity and reliability of optical fiber networks is essential for modern global communication systems, as faults such as fiber cuts, malicious tapping, and connector degradation can result in significant service disruptions and security vulnerabilities. Traditional fault detection methods, such as optical time-domain reflectometry (OTDR), face challenges due to noisy signals, signal attenuation, and reliance on expert interpretation. Additionally, existing machine learning approaches often struggle with generalization across diverse fault types and imbalanced datasets. This paper introduces a novel Hybrid CNN-Ensemble framework, combining convolutional neural networks (CNNs) for deep feature extraction with ensemble learning techniques including XGBoost, Support Vector Machines (SVM), and Random Forest (RF). The framework processes CNN-extracted features from OTDR traces, with predictions aggregated through a meta-classifier. Validated on a comprehensive OTDR dataset covering eight distinct fault categories, the proposed method achieves 99.5% accuracy and an F1-score of 0.99, offering exceptional precision in detecting and classifying fiber faults. The results demonstrate the framework’s ability to improve fault detection performance, streamline fiber monitoring processes, reduce operational costs, and enhance network resilience to both physical attacks and accidental failures.