Creating signed directed graph models for process plants
2010-11-11T11:13:51Z (GMT) by
The identification of possible hazards in chemical plants is a very important part of the design process. This is because of the potential danger that large chemical installations pose to the public. One possible route for speeding up the identification of hazards in chemical plants is to use computers to identify hazards automatically. This will facilitate safe plant design and will avoid late design changes which can be very costly to implement. Previous research at Loughborough has concentrated on developing a model-based approach and an analysis algorithm for automating hazard identification. The results generated have demonstrated the technical feasibility of the approach. This approach requires a knowledge-base of unit models. This library of models describes how different plant equipment behaves in qualitative terms. The research described in this thesis develops a method for creating and testing the equipment models. The model library was previously achieved by an expert writing the models in a format that could be directly used by the system described above. An engineer unfamililar with the system would find this difficult. An alternative method would have been to use an intermediary (a knowledge engineer) to gather information from the engineer and convert it into the system format. This would be expensive. Both methods would take up a lot of the engineer's time. An engineer should be able to enter information personally in order to maintain efficiency and avoid information loss through the intermediary. A front end interface has been built to the system which enables an expert to enter information directly without needing to understand details of the application system. This interface incorporates ideas from the knowledge acquisition field in order to produce a tool that is simple to use. Unit-based qualitative modelling can lead to incorrect or ambiguous inference. The method developed here identifies situations where ambiguities may arise. A new modular approach is presented to overcome this type of problem. This method also presents a technique to verify that the models created are both complete and correct.