A Scalable Specification-Agnostic Multi-Sensor Anomaly Detection System for IIoT Environments

Artikel i vetenskaplig tidskrift, 2020

Advanced sensing is a key ingredient for intelligent control in Industrial Internet of Things (IIoT) environments. Coupled with enhanced communication capabilities, sensors are becoming increasingly vulnerable to cyberattacks, thereby jeopardizing the often safety-critical underlying cyber-physical system. One prominent approach to sensor-level attack detection in modern industrial environments, named PASAD, has recently been proposed in the literature. PASAD is a process-aware stealthy-attack detection mechanism that has shown promising capabilities in detecting anomalous, potentially malicious behavior through real-time monitoring of sensor measurements. Although fast and lightweight, a major limitation of PASAD is that it is univariate, meaning that only a single sensor can be monitored by one instance of the algorithm. This impediment poses serious concerns on its scalability, especially in modernized industrial environments, which typically employ a plethora of sensors. This paper generalizes PASAD to the multivariate case, where a plurality of sensors can be monitored concurrently with little added complexity. This generalization has the evident advantage of offering scalability potential for deployment in future-focused industrial environments, which are undergoing growing integration between the digital and physical worlds.