MUMBAI, India, June 26 -- Intellectual Property India has published a patent application (202641072327 A) filed by Dr. Sunil K. S, Government Engineering College Idukki, Painavu, Kerala; Pradeep Chandran, Government Engineering College Idukki, Painavu; Amritha Remesh, Government Engineering College Idukki, Painavu, Kerala; Veena Roy, Government Engineering College Idukki, Painavu, Kerala; Vidya Roy, Government Engineering College Idukki, Painavu, Kerala; and Vivek D. V, Government Engineering College Idukki, Painavu, Kerala on June 11, 2026, for Topology-Aware Edge-Deployable Intrusion Detection System For Cyber-Physical Attacks In Water Distribution Networks.
Inventors include Dr. Sunil K. S, Government Engineering College Idukki, Painavu, Kerala; Pradeep Chandran, Government Engineering College Idukki, Painavu; Amritha Remesh, Government Engineering College Idukki, Painavu; Veena Roy, Government Engineering College Idukki, Painavu, Kerala; Vidya Roy, Government Engineering College Idukki, Painavu, Kerala; and Vivek D. V, Government Engineering College Idukki, Painavu, Kerala.
The application for the patent was published on June 19, 2026, under issue no. 25/2026.
Abstract: The present invention relates to intrusion detection systems for cyber-physical infrastructures and, more particularly, to a lightweight edge-deployable framework for detecting cyber-physical attacks in water distribution systems. With the increasing digitization and automation of critical infrastructure, water distribution networks have become vulnerable to sophisticated cyber-physical attacks that target sensors, actuators, and control systems. Existing machine learning-based intrusion detection methods often require centralized computation and high processing resources, limiting their deployment in real-time edge environments. The system transforms raw sensor measurements from the hydraulic infrastructure into a topology-aware condensed sensor graph representation that enables efficient zone-level behavioral analysis on resource-constrained edge devices. This trans- formation enables efficient structural reasoning over large-scale infrastructure networks while maintaining low computational overhead suitable for real-time edge deployment. The invention provides a topology-aware intrusion detection system that models the water distribution network as a graph based on its hydraulic layout, where nodes represent junctions, reservoirs, or sensor locations and edges represent pipeline connections. A condensed sensor-focused sub-graph is constructed by retaining only sensor-equipped nodes while preserving structural relationships between them. This reduced graph enables efficient processing while maintaining essential network dependencies. The system employs a spatiotemporal learning architecture in which a Graph Neural Network captures spatial dependencies among sensor nodes while a Gated Recurrent Unit models temporal patterns in sensor measurements. The combined architecture learns the operational behavior of the infrastructure and identifies anomalous patterns associated with cyber-physical attacks. During operation, sensor inputs are dynamically organized into a sub-graph representation and processed by the trained model to enable real-time anomaly detection. To improve adaptability and scalability, the invention incorporates zone-specific lightweight adapter modules that generate localized embeddings representing regional system behavior. These compact modules refine detection for individual zones without increasing computational complexity. The resulting hierarchical detection framework provides efficient anomaly detection, improved attack localization, and low- latency operation suitable for deployment on embedded edge computing devices. The proposed system enables scalable, topology-aware, and computationally efficient protection of water distribution infrastructures against cyber-physical threats.
Disclaimer: Curated by HT Syndication.
