CYPRESS: Cyber-Physical RESilience and Sustainability Dependability Techniques for Instrumented Cyber-Physical Spaces

 

The CYPRESS (CYber Physical RESiliance and Sustainability) project explores techniques for dependability and resilience in instrumented cyberphysical spaces (ICPS) where physical spaces have been instrumented with “intelligence” through heterogeneous sensing, actuation and communication mechanisms. In CYPRESS, we are expoiting an “observe-analyze-adapt” (OAA) architecture in which an ICPS has a model of itself, its objectives, and its effects on the environment; the ICPS achieves dependability objectives through adaptation using runtime application of formal analysis methods. The proposed dependability techniques are cross-layer in nature and range from combining multiple networking and messaging technologies to adaptive sensing and information fusion. We have developed techniques for (i) cross-layer formal modeling of the ICPS environment based on Responsphere/I-Sensorium (a sample ICPS at UC Irvine) and determination of failure modes in ICPS,(ii) characterization of case studies to drive the resilience research, (iii) design of new resilience techniques and formal reasoning/analysis for the designed resilience mechanisms.

To lend focus to the modeling techniques and resilience algorithms, we model the normal functioning of a high-rise campus building instrumented with sensors for a surveillance related application that morphs into a situational awareness application when there is a fire in the building. The objective of the modeling component is to develop a formal framework with executable models that will specify the layered ICPS architecture and application needs, analyze current system state, detect violation of end-to-end dependability requirements, and reason about the validity of adaptations. For example, we study dependability of the ICPS multinetwork; our approach analyzes and verifies carefully selected points in the multinetwork architecture space to address an otherwise intractable problem. We have initiated research in the area of cross-layer virtual sensing and actuations with the aim of overcoming the vulnerabilities introduced by faults, failures and process variability at the architectural and network layers. Techniques to improve reliability, performance, thermal stability, and reduced power and energy consumption applied across different layers of system stack are being studied.

We address resilience at the infrastructure level along two dimensions – the ability to deal with sensor data capture/transfer limitations (resulting in lost information) and the ability to deal with disruptions to the communication infrastructure. To address the former, we are in the process of developing scheduling and load-balancing mechanisms that intelligently determine which sensors to instantiate and what data to transfer. Resilience techniques to understand network failures being studied include (a)exploiting data transfer over multiple networks (b) exploiting mobility (of humans, robots) to ferry data from locations with limited networking to points where network connectivity is adequate the captured information. For example, we are developing a multinetwork management platform (MINA - Multinetwork INformation Architecture), a middleware system that collects network state information from heterogeneous networks (overcoming lower level protocol related constraints) and provides management functions to address resource provisioning, fault analysis and power management across multiple networks.

Our work on information resilience uses event semantics to enable cooperative actuation of multiple devices (e.g. cameras in a video surveillance use case) and exploits context semantics to improve the accuracy of information extraction using entity resolution approaches. We develop adaptive cooperative actuation techniques that effectively balances the tradeoffs between event accuracy and event capture through a semantics-based lightweight, real-time scheduler. To reduce information uncertainty by implementing techniques for data cleaning and entity resolution, we connect the problem of person identification in video data with the problem of entity resolution that is common in textual data and use a graph-based entity resolution framework called RelDC that leverages relationships among various entities for reliable disambiguation. CYPRESS researchers are organizing a focused workshop on Reliable Cyberphysical Systems WRCPS 2012 (http://www.ics.uci.edu/~dsm/wrcps2012) to be coheld with the IEEE Symposium on Reliable Distributed Systems in October 2012 in Irvine CA. 96

Award Number: 106356