Representing Operational Modes for Situation Awareness

Operating complex plants is an increasingly demanding task for human operators. Diagnosis of and reaction to on-line events requires the interpretation of real time data. Vast amounts of sensor data as well as operational knowledge about the state and design of the plant are necessary to deduct reasonable reactions to abnormal situations. Intelligent computational support tools can make the operator's task easier, but they require knowledge about the overall system in form of some model. While tools used for fault-tolerant control design based on physical principles and relations are valuable tools for designing robust systems, the models become too complex when considering the interactions on a plant-wide level. The alarm systems meant to support human operators in the diagnosis of the plant-wide situation on the other hand fail regularly in situations where these interactions of systems lead to many related alarms overloading the operator with alarm floods. Functional modelling can provide a middle way to reduce the complexity of plant-wide models by abstracting from physical details to more general functions and behaviours. Based on functional models the propagation of failures through the interconnected systems can be inferred and alarm floods can potentially be reduced to their root-cause. However, the desired behaviour of a complex system changes due to operating procedures that require more than one physical and functional configuration. In this paper a consistent representation of possible configurations is deduced from the analysis of an exemplary start-up procedure by functional models. The proposed interpretation of the modelling concepts simplifies the functional modelling of distinct modes. The analysis further reveals relevant links between the quantitative sensor data and the qualitative perspective of the diagnostics tool based on functional models. This will form the basis for the ongoing development of a novel real-time diagnostics system based on the on-line adaptation of the underlying MFM model.