CRII: CPS: Data-Driven Cascading Failure Abstraction and Vulnerability Analysis in Cyber-Physical Systems

Santa Clara University, Santa Clara CA

Investigators

Abstract

The goal of this proposal is to establish a framework for cascading failure abstraction and vulnerability analysis in Cyber-Physical Systems (CPSs), empowered by data. CPSs are critical to modern society, however, they are vulnerable to attacks and failures. The failures in CPSs are more destructive because of cascading failure, which means that the failure of a part of the system can cause failure in the rest of the system and result in more severe damage. However, analysis of CPS vulnerability involving cascading failure is extremely challenging, mainly because 1) it’s hard to theoretically analyze the various physical processes happen in a cascade and 2) local diffusion models applied to the CPS network cannot capture the global impact of cascades. Using simpler cascade models derived from data as media, it is possible to have a deeper understanding of how CPSs are vulnerable to cascading failure. CPSs are gaining popularity and there is an urgent need to enhance its security, hence the proposed work will greatly benefit the society and of national interest. The project will provide opportunities for undergraduate students, underrepresented minority groups and women to research in some of the society’s most concerned fields like machine learning and security. Also, the outcomes of this work will be introduced in courses for undergraduate and graduate students and integrated into STEM outreach programs for K-12 students. The proposal goals will be achieved via four major tasks. (i) Develop simple meta-cascade models that preserve key features of cascades. (ii) Quantitatively measure vulnerability in CPSs with general damage functions defined on failed components and multiple meta-cascade models, as well as design protection schemes. (iii) Develop efficient sampling based dynamic approximation algorithms, which can run efficiently in billion-scale networks and can be stopped at any time with theoretical guarantees. The algorithms are widely applicable to optimization problems in large-scale networks and are essential in supporting the first two tasks. (iv) Implement and evaluate the proposed models and algorithms. The innovations in meta-cascade models can benefit research in CPS security, while the sampling-based algorithms can be a powerful new tool for a wide class of optimization problems in large-scale networks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.