Research Thrusts

Welcome to

Complex Systems Failure (CSf)

A Multidisciplinary University Research Initiative (MURI)

esearch Thrusts

	Research Overview and Approach
	Mathematical Modeling of Fault Generation and Propagation (Dr. A. Ray)
	Characterization of Pathological Behavior (Dr. D. P. Siewiorek)
	Damage Mitigating Control (Dr. A. Ray)
	System Dependability (Dr. V. Phoha)

Research Overview and Approach

This project addresses the fundamental research issues of characterizing and mitigating (or circumventing) failures in complex informational systems for mission-critical defense applications. We define complex dynamic systems to consist of heterogeneous structured subsystems which dynamically interact at multiple levels of interactions. Structured subsystems utilize complex combinations of multiple system resources to accomplish tasks or services. A physics-based model of the essential dynamics of pathological behaviors in subsystem interactions is envisioned so that time-critical applications involving multiple resources, with individually evolving needs, can be analyzed and controlled within a distributed information infrastructure. The model is designed to dynamically utilize available resources to achieve stability under conditions of uncertainty, environmental noise and partial information. Essential features of this model and its implementation are:

	Co-existence and inter-operation of independently evolving services, with possibly inconsistent and conflicting objectives and resource requirements.
	To each resource provided by the system is associated a service agent that controls the utilization of this resource through interactions with other agents or with the environment.
	An application is a meta-level composition of services that can be decomposed into process algebraic expressions of elementary services that utilize a single resource. This concept is similar to a high-level instruction being decomposed into machine instructions for execution on parallel machines. All applications are considered ill-posed because their execution-level dynamics are not yet defined.
	Some system components could be mobile.

Our aim is to formulate analytical models of the higher level dynamics of subsystem interactions: (i) to predict emerging pathological behavior from time series observations of events (classified from sensor data) and interactions, and (ii) to formulate adaptive mechanisms to circumvent or mitigate the effects of pathological behavior.

In many ways, the macroscopic behavior of the aggregate system emerges from interactions between individual components. The evolutionary physics of these systems has generated great interest in the continuous case [Alligood 96, Nicolis 77, Haken 78] for studying physical phenonmena like the flow of gases, flexible structures, chemical reactions and biological systems. More recently discrete event models for distributed control in human-engineered systems have also been studied for military operations [Phoha 99], computer network management [Lee 93], mechanical and aerospace systems [Zhang 00], and Internet applications [Leland 94, Grossglauser 99].

The search for fundamental principles of fault tolerance in human-engineered complex systems is very new. Our search for analytical notions of pervasive fault tolerance is based on the fact that many biological and chemical systems exist where microscopic flux and chaos are offset by macroscopic order. We will draw on (i) the extant scientific knowledge-base on the physics of individual failure and (ii) the relatively sparse knowledge-base on complex systems behaviors to discover the fundamental principles for pervasive fault tolerance in human-engineered systems.

Top