How to do knowledge acquisition without completely annoying your expert

A central problem in producing any expert system is the elucidation of the knowledge from the expert. However, there is a large source of knowledge that is often overlooked by the knowledge engineer, and that is information direct from the process that you are trying to model. Unfortunately this data, when in its raw form, is unusable by a standard expert system; what is needed is some way of extracting the useful information, the patterns, from the data. In other words some form of data mining needs to be performed. Database mining systems are useful for detecting trends in large quantities of data, but they function best with some sort of guidance. This is the role we suggest for the expert. In such “hybrid” systems, the system does most of the knowledge acquisition itself, but the expert determines what sort of knowledge should be acquired and from which sources. The TIGON system is being developed to detect and diagnose faults in an industrial gas turbine engine. The aim of the TIGON project is to produce a similar a set of knowledge bases as produced manually in the TIGER project. An additional aim is to modify the TIGON-produced knowledge bases so that they are applicable to further turbine systems. To this end, we have developed a methodology that enables TIGON to mine the data that has been routinely collected by the online computer while the turbine is operating