Detecting probability footprints of sensor deception attacks in supervisory control

Sensor deception is a class of attacks in control systems where an attacker manipulates sensor readings to cause damage to the system. In this work, we investigate the problem of designing better and faster intrusion detection systems against sensor deception attacks. We study this problem in the context of stochastic supervisory control theory using the notion of ∈-safety detection. The ∈-safety notion ensures that a sensor deception attack can be detected due to changes in the probabilistic behavior in the control system, i.e., it leaves a probability footprint. We provide necessary and sufficient conditions to verify if a system is ∈-safe in polynomial-time complexity improving the current state-of-the-art exponential-time complexity.