An Edge-Enabled Multimodal Cyber-Physical System for Near-Real-Time Intrusion Detection in Fiber-Optic Networks

Fiber-optic backhaul and access networks remain vulnerable to excavation, vandalism, cable pulling, and other physical disturbances, particularly in remote deployments where continuous cloud connectivity and expensive optical interrogators are impractical. This paper presents FOC-IDS, an edge-enabled multimodal cyber-physical intrusion detection system designed for low-infrastructure environments. The proposed architecture integrates vibration, acoustic, temperature, and humidity sensing with an ESP32-based support vector machine (SVM), confidence-aware node consensus, and GSM-SMS alerting. Unlike conventional OTDR- or DAS-based approaches, FOC-IDS emphasizes offline-first operation, low deployment cost, and lightweight edge inference. Field data collected in Minna, Nigeria, were used to train and evaluate the system under realistic environmental conditions. On a held-out test set, FOC-IDS achieved 98.03% accuracy, 1.00 precision for the intrusion class, 0.95 recall, and an AUC-ROC of 0.992, with a mean end-to-end response latency of 6.45 s. Consistent with the measured latency, the system is characterized as near-real-time rather than strict real-time. Ablation experiments further demonstrate that full multimodal fusion outperforms unimodal and dual-modal configurations, while a humidity-adaptive decision rule improves intrusion recall under heavy rain by 9.1 percentage points. The paper additionally discusses reproducibility, deployment constraints, cross-site generalization limitations, and alignment with IEC 62443-oriented security principles. Overall, the proposed system provides a practical and standards-aware framework for protecting fiber-optic infrastructure in connectivity-constrained environments.