Introducing dynamics in a fault diagnostic application using Bayesian Belief Networks

Fault diagnostic techniques are required to determine whether a fault has occurred in a system and to identify the component failures that may have caused it. This task can be complicated when dealing with complex systems and dynamic behaviour, in particular, introduces further difficulties. This paper presents a method for fault detection on dynamic systems using Bayesian Belief Networks (BBNs). Possible trends are identified for the variables in the systems that are monitored by the sensors. Fault Trees (FTs) are built to represent the causality of the trends and these are then converted into BBNs. The networks developed for different sections are connected together to form a unique concise network. For a combination of sensors which deviate from the expected trends, calculating the updated probability enables a list of potential causes for the system scenarios to be obtained. A simple water tank system has been used to validate the method.