Remotely inferring device manipulation of industrial control systems via network behavior

This paper presents preliminary findings on a novel method to remotely fingerprint a network of Cyber Physical Systems and demonstrates the ability to remotely infer the functionality of an Industrial Control System device. A monitoring node measures the target device's response to network requests and statistically analyzes the collected data to build and classify a profile of the device's functionality via machine learning. As ICSs are used to control critical infrastructure processes such as power generation and distribution, it is vital to develop methods to detect tampering. A system employing our measurement technique could discover if an insider has made unauthorized changes to a device's logic. Our architecture also has advantages because the monitoring node is separate from the measured device. Our results indicate the ability to accurately infer (i.e., using a tunable threshold value) discrete ranges of task cycle periods (i.e., CPU loads) that could correspond to different functions.