Distributed Diagnosis of Networked, Embedded Systems
James Kurien,Xenofon D. Koutsoukos,Zhao Feng
2002-01-01
Abstract:Networked embedded systems are composed of a large num- ber of physically distributed nodes that interact with the phys- ical world via a set of sensors and actuators, have their own computational capabilities, and communicate with each other via a wired or wireless network. Monitoring and diagnosis for such systems must address several challenges caused by the distribution of resources, communication limitations, and node and link failures. This paper presents a distributed di- agnosis framework that exploits the topology of a physical system to be diagnosed to limit inter-diagnoser communica- tion and compute diagnoses in an anytime and any informa- tion manner, making it robust to communication and proces- sor failures. The framework adopts the consistency-based di- agnosis formalism and develops a distributed constraint sat- isfaction realization of the diagnosis algorithm. Each local diagnoser first computes locally consistent diagnoses, tak- ing into account local sensing information only. The local diagnosis sets are reduced to globally consistent diagnoses through pairwise communications between local diagnosers. The algorithm has been successfully demonstrated for the di- agnosis of paper path faults for the Xerox DC265 printer. the network dynamically and nodes of a previously unseen type might join in their place. In this paper, we consider how we might apply techniques from model-based diagnosis to these types of problems. In general, traditional model-based techniques are centralized. They assume that the diagnostic algorithm is run on a sin- gle processing unit that has access to observations from all sensors in the physical plant. In the next two sections of the paper, we briefly discuss centralized, model-based tech- niques and discuss how they cause scalability, robustness and reconfigurability problems if employed directly on net- worked, embedded systems. We then present a set of use- ful properties for diagnostic algorithms for such systems. In the fourth section, we present a simple formulation for diagnosis of discrete, distributed systems in order to mo- tivate discussion and map the formulation onto distributed constraint satisfaction and distributed constraint optimiza- tion. We next propose an algorithmic framework for dis- tributed diagnosis that operates in an anytime manner and is robust to communication and processor failures. We dis- cuss the communications requirements for the framework and compare performance results for one instantiation of the distributed diagnosis framework against a centralized diag- noser. In the related work section, we discuss why exist- ing distributed constraint satisfaction and optimization algo- rithms are not well suited for distributed diagnosis of net- worked, embedded systems. We finally discuss two open areas for future work. The contributions of this paper are that it illustrates the interesting features of networked, em- bedded systems that make them challenging for traditional model-based diagnosis techniques, it presents a simple for- mulation of the distributed diagnosis problem for these type of systems and relates it to distributed constraint satisfaction and optimization, it presents a class of robust, anytime al- gorithms for performing diagnosis, and it illustrates prelim- inary diagnostic results on a model of a real physical system with comparisons to an existing centralized diagnoser.