Quantitative dependability and interdependency models for large-scale cyber-physical systems

Cyber-physical systems link cyber infrastructure with physical processes through an integrated network of physical components, sensors, actuators, and computers that are interconnected by communication links. Modern critical infrastructures such as smart grids, intelligent water distribution networks, and intelligent transportation systems are prominent examples of cyber-physical systems. Developed countries are entirely reliant on these critical infrastructures, hence the need for rigorous assessment of the trustworthiness of these systems. The objective of this research is quantitative modeling of dependability attributes -- including reliability and survivability -- of cyber-physical systems, with domain-specific case studies on smart grids and intelligent water distribution networks. To this end, we make the following research contributions: i) quantifying, in terms of loss of reliability and survivability, the effect of introducing computing and communication technologies; and ii) identifying and quantifying interdependencies in cyber-physical systems and investigating their effect on fault propagation paths and degradation of dependability attributes. Our proposed approach relies on observation of system behavior in response to disruptive events. We utilize a Markovian technique to formalize a unified reliability model. For survivability evaluation, we capture temporal changes to a service index chosen to represent the extent of functionality retained. In modeling of interdependency, we apply correlation and causation analyses to identify links and use graph-theoretical metrics for quantifying them. The metrics and models we propose can be instrumental in guiding investments in fortification of and failure mitigation for critical infrastructures. To verify the success of our proposed approach in meeting these goals, we introduce a failure prediction tool capable of identifying system components that are prone to failure as a result of a specific disruptive event. Our prediction tool can enable timely preventative actions and mitigate the consequences of accidental failures and malicious attacks --Abstract, page iii