Probabilistic diagnosis

Methods for identifying bus-type faults, instruction faults, switching system faults, and certain architectural faults are described. These methods are particularly applicable in very-large-scale computational assemblies. The approach used for diagnosis is probabilistic, in the sense that the data sets used contain a partial indication of the fault location and are in some sense random. Use of a sequence of such data sets can progressively isolate system faults. Mean values and associated variance figures for the number of faulty data sets required for a diagnosis in each of the methods described are given. Considerations involved in applications of this diagnostic process are discussed