Digital Chemical Engineering (Mar 2023)
Development of directed randomization for discussing a minimal security architecture
Abstract
Strategies for mitigating the impacts of cyberattacks on control systems using a control-oriented perspective have become of greater interest in recent years. Our group has contributed to this trend by developing several methods for detecting cyberattacks on process sensors, actuators, or both sensors and actuators simultaneously using an advanced optimization-based control strategy known as Lyapunov-based economic model predictive control (LEMPC). However, each technique comes with benefits and limitations, both with respect to one another and with respect to traditional information technology and computer science-type approaches to cybersecurity. An important question to ask, therefore, is what the goal should be of the development of new control-based techniques for handling cyberattacks on control systems, and how we will be able to benchmark these as “successful” compared to other techniques to drive development or signal when the research in this direction has reached maturity. In this paper, we propose that the goal of research in control system cybersecurity for next-generation manufacturing should be the development of a security architecture that provides flexibility and safety with lowest cost, and seek to clarify this concept by re-analyzing some of the security techniques from our prior work in such a context. We also show how new methods can be developed and analyzed within this “minimum security architecture” context by proposing a technique which we term “directed randomization” that may require less sensors to be secured in a system than some of our prior methods, potentially adding flexibility to the system while still maintaining security. Directed randomization seeks to utilize the existence of two possible stabilizing inputs at every sampling time to attempt to create a challenge for an attacker for setting up an arbitrary sensor attack policy without being detected within a finite number of sampling periods. We discuss benefits and limitations of this technique with respect to our prior cybersecurity strategies and also with respect to extended versions of these prior concepts, such as image-based control and distributed control, to provide further insights into the minimum security concept.
