Probabilistic Algorithm for System Level Self-diagnosis

System-level self-diagnosis is one of the most important tasks in the field of computer science. In this paper, we present the results of the research on how to increase the credibility of the results of system diagnosis by way of merging two methods of system-level self-diagnostic (traditional and unconventional). As distinct from traditional system level self-diagnosis, unconventional method of system diagnosis can deal with arbitrary testing assignments and can be applied to heterogeneous systems. The diagnosis problem consists of determining the location of faults in the system (i.e., determining faulty units). In traditional system level self-diagnosis, such diagnosis problem can be defined as finding the necessary and sufficient conditions for a system testing assignment that should be satisfied to achieve a given level of diagnosability given a fault model and an allowable family of fault sets. For solving the diagnosis problem the appropriate diagnosis algorithms should be developed. Before designing a diagnosis algorithm it is needed to adopt the strategy that is suitable for the particular complex system. Among the possible diagnosis strategies, such as unique, sequential, excess and probabilistic, the probabilistic strategy was chosen. Based on this strategy, the diagnosis algorithms were designed. The results (credibility) of the algorithms that follow the probabilistic diagnosis strategy can be improved. For this purpose, some elements of unconventional system level self-diagnosis are used in the algorithm design. Short description of unconventional system level self-diagnosis is presented in this paper. The obtained results of improved diagnosis allow revealing the functional dependence of the credibility of diagnosis results on the values of system and testing parameters.