Process system fault source tracing based on Bayesian networks

Abstract

Process systems are different from discrete manufacturing systems in that they are composed of many interlocking subsystems that consist of various tightly coupled units. Hence, whenever a small unit of a subsystem functions badly, it could influence or cause the whole system to function abnormally. Under this circumstance, how to locate abnormal or failed units will be a very formidable task. By introducing Bayesian networks into the tracking of fault sources in process systems, a new method of fault source tracing is proposed, and a model is set up accordingly. To guarantee the accuracy of the modeling process, a series of rules which must be abided by are defined. And, a mapping between the Bayesian network and the units of a process system is developed. Furthermore, to make full use of this new Bayesian network model, its probability characteristics and the problem-solving method exploiting it are expanded and investigated in depth. Additionally, a complete reasoning for the fault source tracing based on the Bayesian network is described. Finally, an example is provided to demonstrate the modeling and reasoning processes, and thereby verifies the practicality and validity of this model in tracing abnormalities in process systems.