An Analytical Frameowrk

Motivation

Cyber-physical system (CPS) was coined in 2006 by the National Science Foundation (NSF) to refer to the tight conjoining of and coordination between computational and physical resources. The closed-loop interaction between cyber and physical components increase the risk of cyber attacks leading to physical catastrophes.

While important results have been generated in

• vulnerability analysis,
• anomaly detection,
• resilient control,
• estimation,
  there is a lack of analytical bridges among these multidisciplinary techniques due to the modeling gap between the cyber and physical worlds.
  

This lack of analytical bridge hinders further improvement in CPS security. For example,

• the physics-driven resilient estimation designs suffers from a fundamental limitation that requires more than half of sensors to be safe;
• the anomaly detection and isolation approaches require extensive redundancy in sensor deployment.
  

Example cyber threats in the history:

State-of-the-art

Developing concurrent physics-driven and data-driven learning approaches to advance the resiliency of cyber-physical systems in ways that are unattainable using either approach alone.

• Tools:

• Proposed framework

The concurrent model resilient estimation and control (CM-REC) framework, illustrated above, composes of

• Automated attack generator (AAG)
• Attack detector (AD)
• Pruning algorithm (PA)
• Resilient estimator (RE)
  AAG explores the vulnerability space of CPS and generate feasible attack datasets from noise. AD is trained with the generated attacks, and the coupling of AAG and AD can explore the entire vulnerability space of CPS. PA is a statistical algorithm analytically removing possible false positive (FP) cases, where the detector labels the information source as safe, but it is not, in the detection conclusion. Thereafter, the attack detection’s precision is improved. A weighted RE is used to utilize the detection prior in the estimation design. The coupling of PA and the weighted RE provides a quantitative bridge between the detection precision and the estimation error bound. In addition, our recent results indicate that this framework maintain resiliency even when 60% of sensors are attached. Please see other articles for full details of each techniques.