MACHINE FAILURE ISOLATION USING QUALITATIVE PHYSICS

摘要

A Qualitative Reasoning System performs machine failure isolation by generating a plurality of pending hypotheses (108) which assume the failure of one or more machine components and by testing each of the hypotheses by propagating actual values for machine parameters through confluence equations for the hypothesis being tested. Variables associated with hypotheses of a qualitative physics model are grouped into type I variables and type II variables, wherein any possible value measured for a type I variable is guaranteed to render at least one hypothesis inconsistent and some, but not all, possible values measured for a type II variable will render at least one hypothesis inconsistent. A set of confluences is deemed to be inconsistent by determining that there is no set of values which can be assigned to all unknown variables of the confluences which result in a consistent set of predictions. The value of an unknown variable is determined by iteratively setting the variable to each of its possible values and, for each iteration, propagating the known variables and the value of the unknown variable through confluences of the model. A model instantiator converts computer workstation user input into usable model information and provides other optimizations which reduce the number of confluences and variables of the model.