Covert Cognizance: A Novel Predictive Modeling Paradigm

Abstract

Can predictive models develop cognizance or awareness of how they have been used? Can models detect if they are being manipulated or executed in nonauthorized manners? Can a software track information propagation through its subroutines to improve execution efficiency? Can this be achieved in a covert manner, i.e., avoiding the use of additional variables, additional lines of code, and conventional logging files, and instead rely directly on the physics being simulated to develop the required cognizance? Achieving these goals under the looming threat of insiders is considered an open challenging problem. This paper introduces a new modeling paradigm to covertly develop cognizance that is of critical value when predictive software is used in both adversarial and nonadversarial settings. Given the wide range of applications possible with this new modeling paradigm, the paper will focus on introducing the mathematical theory and limit the initial demonstration to a physics-based model of a nuclear reactor. This model describes a representative industrial control system of a nuclear reactor model containing two coupled subsystems: a heat-producing core and a steam generator. The goal is to demonstrate how each subsystem physics model can remain cognizant of the state of the subsystem. The proposed methodology will provide communication solutions for future reactor technologies to enable advanced reactor control and remote reactor operations.

Keywords:

• Cybersecurity
• predictive science
• nuclear reactor control
• reduced-order modeling
• software cognizance
• embedded intelligence

Acknowledgments

This work has received support from multiple sources, including the Purdue University School of Nuclear Engineering, U.S. Department of Energy Office of Nuclear Energy, and U.S. Army Research Laboratory.