Integrating statistics, numerical analysis and dependency-recording in model-based diagnosis

Abstract

Research on model-based diagnosis of technical systems has grown enormously in the last few years, producing new basic tools, new algorithms and also some applications. However, the majority of research has dealt with systems described by variables ranging in discrete domains (e.g., digital circuits), and only few attempts have been made at applying such techniques to continuous domains. Continuous systems are characterized by additional problems, such as the unavoidable sensor errors and the need for using more complex models which may consist of simultaneous non-linear equations. The distinctive feature of the approach we present in this paper is the integration of techniques well known in the field of numerical analysis and statistics (e.g., the solution of non-linear systems and the error propagation) with a dependency-recording technique based on ordering the equations and the variables of the model.