Automated Anomaly Detection in Industrial Control System (ICS) Network Traffic via Multi-Modal Graph Neural Network Fusion

**Abstract:** This research proposes a novel approach to detecting anomalies within Industrial Control System (ICS) network traffic by fusing multiple data modalities—packet metadata (protocol, source/destination IP, port), time-series feature extraction from flow data (inter-arrival times, packet sizes), and textual representations of control commands—using a multi-modal Graph Neural Network (MGNN) architecture. This fusion strategy surpasses the limitations of single-modality approaches, enabling early detection of sophisticated attacks and operational deviations that would otherwise go unnoticed. The system is designed for immediate commercial deployment, leveraging established deep learning techniques and demonstrating superior performance across established ICS benchmark datasets.