Hardware-in-the-loop Based Investigations of False Data Injection Attacks on Smart Water Systems Using Autoencoders: A Real-time Validation Framework

Abstract Submission: Cyber accessibility for the monitoring and control of smart water distribution systems (SWDS) is facing emerging cyber-physical attacks, deteriorating systems’ security with increased intrusion sophistication. Overcoming such attacks requires in-situ investigations of cyberattack consequences in the SWDS to assess their vulnerabilities and required remedial approaches. The available standard datasets are simulated in a software environment, hinting at the scarcity of real-world datasets. Acknowledging the lack of such datasets, this paper presents real-time consequences of false data injection attacks (FDIA) on an experimental testbed, depicting a real-world SWDS, using hardware-in-the-loop (HIL) testing. The testbed comprises a reservoir, overhead tank, and smart devices (such as flow meters, pressure sensors, controllable valves, and pumps) operating in a real-time LabView environment framing its supervisory control and data acquisition (SCADA). FDIA investigations are performed by gradual manipulation of the main pump’s controller resulting in cutting the water supply to the entire network. The series of cyberattack intrusions are investigated in real time to establish a real-world FDIA dataset. Three datasets are obtained including training dataset (clean events only), validation, and test datasets containing clean and FDIA events launched at different times of the day. The resulting datasets of clean and malicious events facilitate the development of a robust FDIA detector employing Autoencoders, correctly detecting all attack events with high classification scores. The comprehensive evaluation of FDIA impacts on the testbed and its detection experimentally validates the effectiveness of the proposed cybersecurity frameworks for real-world applications, while providing a benchmark for future cybersecurity studies.

Learning Objectives/Expected Outcome (Optional) : 1. A real-world cyberattack dataset for future studies
2. A benchmark case study of false data injections in the water testbed.
3. A robust cyberattack detection framework for real-world applications.