Hardware-in-the-Loop Based Investigations of a Multi-Modal Detection Framework for Cybersecurity and Operational Failures in Smart Water Systems: A Real-Time Validation Framework

Smart water distribution systems face growing risks from cyber-physical attacks, pipe leaks, and sensor malfunctions, which can compromise system security and lead to water loss or inaccurate control actions. The scarcity of real-world datasets for such events, with most data derived from simulations, highlights the need for realistic data. This study introduces a multi-modal detection framework, trained and validated using real-time datasets from an experimental testbed designed to replicate a real-world SWDS. The testbed includes a reservoir, overhead tank, flow meters, pressure sensors, controllable valves, and pumps, managed in a real-time LabView environment to simulate supervisory control. This hardware-in-the-loop setup allows realistic testing of cyberattacks, including pump manipulation to disrupt water supply. Five datasets were generated, covering clean operations, cyberattacks, sensor failures, and leaks. Using autoencoders, the framework achieved high detection accuracy (F1 > 0.94) across all events, enhancing SWDS security and reliability.