Secure Decision Making and Inference in Critical Systems

Abstract

The use of digital components in critical systems creates vulnerabilities. Artificial Intelligence (AI) and Machine Learning (ML) can be used towards the monitoring of a system, because of their ability to handle big volumes of data. Specifically, the physical nature of a category of these systems – the cyber physical systems, (CPSs) – permits the embedding of the concept of inference and reinforcement learning. According to this, a system can perform decision making by selecting actions, which later provide certain rewards. This idea is implemented on a simple nuclear system, to determine whether the system has the ability to adjust to a specific power level by making decisions about other variables, such as control rod heights. Some preliminary results show that the system is able to adjust its state after some time has passed.