Supporting model based diagnosis with evidence

The `traditional´ approach to diagnosis in plant and machinery has been to identify alarm conditions and present to the operator a indication of the alarm state when it arises. In certain pathological cases the flow of alarms can be sudden and in such numbers that the operators quickly become overloaded. However operators are frequently very successful if given sufficient time to assimilate the information. An operator´s understanding of the machinery processes may be limited to some prior experience and not the principles on which it operates. By resorting to a description of the process at the level of governing principles, there is a clearer opportunity to find and correct the problem. The author presents the framework for using qualitative models in a framework where evidence from external tests can be introduced. The framework takes the form of a series of knowledge sources in a knowledge-based supervisory system