Predicting faults with real-time diagnosis

The authors discuss a strategy for implementing real-time expert systems when the expert system hardware-software combination is too slow at first analysis. A major aspect of the system is its ability to predict faults. The techniques described have been fully implemented on a large-scale real-time expert system for British Steel at Ravenscraig in Scotland. This application is discussed in detail. The authors outline the requirement for a real-time system and how real-time analysis has been provided in spite of the fact that the computer system and expert system are notionally too slow for real-time input. The heart of the real-time capability is a result of planning ahead for various types of data analysis and providing adequate history management so that the analysis is available at the appropriate time. The authors also discuss the innovative strategy for allowing the system to predict faults based on the actual plant performance