A consistency checker for a fuzzy diagnosis system applied to warm rolling-mills in steelmaking plants

Technical diagnosis is often confronted with machines that still work but should possibly be repaired to increase efficiency or to avoid general failure. Post-mortem repair strategies are usually too late and tend to increase costs. Prophylactic maintenance of machinery at regular intervals does not lead to unexpected breakdowns but is equally costly. The best strategy would be repairing or replacing defective equipment at a certain optimal point in time usually shortly before imminent failure, the latter being in general difficult to predict correctly. However, it is possible to monitor and capture various parameters coming from the equipment through sensors and to establish some connection between these and the danger of such failure. Nevertheless, given a diagnosis problem, it is usually difficult to choose a priori parameters describing how the available data should be considered. Counterintuitive small changes in the parameters can have chaotic effects on maintenance decisions. We propose a method to select a diagnosis model corresponding to the equipment at hand. The method basically allows a consistent tuning of parameters. Some first analysis indicates that our method makes fuzzy diagnosis systems ready for the real-world.