Model Based Diagnosis Using Directed Graph

Abstract

This paper describes our approach to propose a model-based fault diagnostic method using a directed graph. The model is composed of three sub-models a failure propagation model, a component model and a state transition model. A failure propagation model represents causal relationship among process variables. Here, a pressure low variable, a temperature high variable, and others are represented as nodes, and their causalities are represented as links. A component model represents components and parts structure of the target process. Each faulty status (pipe clogged, pump failure, and so on) is represented as a node. A state transition model is introduced to represent different process operation phase like “start-up” and “steady state.” Each state is represented as a node, and their phase transition is represented as a link. The basic concepts of diagnostic algorithms are as follows : (1) input information on abnormal process variables, (2) find primal failure causes, (3) filter the result. In this paper, model components, modeling method, and diagnostic algorithm are described. The software environment for the models is also discussed.