Node Grouping in System-Level Fault Diagnosis

With the popularization of network applications andmultiprocessor systems, dependability of systems has drawn considerableattention. This paper presents a newtechnique of node grouping for system-level fault diagnosis to simplifythe complexity of large system diagnosis. The technique transforms acomplicated system to a group network, where each group may consist ofmany nodes that are either fault-free or faulty. It is proven thatthe transformation leads to a unique group network toease system diagnosis. Then it studies systematically one-step t-faultsdiagnosis problem based on node grouping by means of the concept ofindependent point sets and gives a simple sufficient and necessarycondition. The paper presents a diagnosis procedure for t-diagnosablesystems. Furthermore, an efficient probabilistic diagnosis algorithmfor practical applications is proposed based onthe belief that most of the nodes in a system are fault-free. Theresult of software simulation shows that the probabilistic diagnosisprovides high probability of correct diagnosis and low diagnosis cost, andis suitable for systems of any kind of topology.