AICrit: A Design-Enhanced Anomaly Detector and Its Performance Assessment in a Water Treatment Plant

Abstract

Critical Infrastructure Security Showdown 2021—Online (CISS2021-OL) represented the fifth run of iTrust’s international technology assessment exercise. During this event, researchers and experts from the industry evaluated the performance of technologies designed to detect and mitigate real-time cyber-physical attacks launched against the operational iTrust testbeds and digital twins. Here, we summarize the performance of an anomaly detection mechanism, named AICrit, that was used during the exercise. AICrit utilizes the plant’s design to determine the models to be created using machine learning, and hence is referred to as a “design-enhanced” anomaly detector. The results of the validation in this large-scale exercise reveal that AICrit successfully detected 95.83% of the 27 launched attacks. Our analysis offers valuable insights into AICrit’s efficiency in detecting process anomalies in a water treatment plant under a continuous barrage of cyber-physical attacks.

Keywords

• anomaly detection
• critical infrastructure
• industrial control systems
• critical infrastructure security showdown
• water treatment plant