A dynamic fault localization algorithm using digraph

Abstract

Analyzed here is a dynamic learning fault localization algorithm based on directed graph fault propagation model and feedback control. Input and output of the algorithm are named as fault and hypothesis respectively. Because of the complexity and uncertainty of fault and symptom, it's difficult to accurately model the relationship of them in probabilistic fault localization. Fault localization algorithm depends on the prior specified model, and the parameter and structure of model is approximate correct and often differ from the real situation. So we propose DMCA+ algorithm which has 3 features: reduce the requirement for accuracy of initial conditions; statistically learn to automatically adapt the probability distribution of fault occurrence while localizing fault; generalize the MCA+ algorithm of no feedback. The feedback learning is similar with proportional adjusting of PID control, but increment is sensitive to detection rate because little increment adjusts output too slowly and big will result in a large number of error hypotheses. The simulation results show the validity and efficiency of dynamic learning under complex network. In order to promote detection rate, optimizing measures are also discussed.