Industrial plants have become increasingly complex with the introduction of more and more computer based control systems and large amounts of sensors throughout a system. While fault diagnosis and fault-tolerant control are employed in many systems to compensate for failures dynamically, human operators are responsible for the diagnosis and appropriate reaction to failures on a plant-wide level. Operators rely on alarm systems, which report abnormal states of systems throughout the plant.
The alarm systems meant to aid the human operators in identifying the plant-wide situation regularly fail in situations where the interactions of systems in the plant lead to many related alarms at once, overloading the operator. Functional modelling, specifically Multilevel Flow Modelling (MFM), has been shown to provide the means to reason about related failures in a system and their root cause as well as potential consequences. This reasoning allows to reduce the number of alarms by considering failure paths.
This project aims at integrating on-line event analysis of sensor data with the framework of MFM to enable more intelligent alarm-systems for plant-wide diagnosis.