Bond Graphs and the Semi-Qualitative Encoding of the Faulty Behavior in Conductive Flow Systems

Abstract

The fault diagnosis in industry is difficult due to the imprecise and the incomplete information related to faults, as well as to the uncertain relations between the observed variables at fault. The faulty model of the target system is not available; only the human diagnostician knowledge on the faulty behavior of the system may be used, i.e. the shallow knowledge (as cases) and the deep knowledge (as qualitative relations between variables), all represented in a linguistic manner on the manifestations and the physical laws in the target system behavior. The paper presents an encoding approach of the power variables (pressure like and flow-rate like variables) in conductive flow systems, as fuzzy partition related to normal and abnormal situations. While in the knowledge acquisition phase some variables have known partition, but many other do not, the paper proposes a method to determine the unknown partitions from the known ones based on conductive flow laws in a qualitative manner, to preserve the semantic consistency of the derived partitions.