Expert system; fuzzy logic; maintenance; steel production; quality; scheduling; continuous caster; AI and knowledge-based systems; uncertainty management.
One strategy when scheduling is to minimize tundish changes in casting sequences to extend the utilization time of each tundish. This time is mainly limited by wear. Another scheduling strategy is to assign heats requiring a quality separation in form of a tundish change as the last heat before the expected end of the tundish's utilization time. Both strategies require a good" knowledge of the tundish's life-expectancy. Human scheduling experts normally play safe and assume a life-time of 240 minutes for one tundish. Problems arise when the real life-time is shorter than 240 minutes, or otherwise exceeded: This results in quality degradations or even the need to reschedule remaining heats, with possibly far-reaching consequences for delivery dates to customers. Additionally, in case of interruptions for unrelated reasons, e.g. machine breakdowns, knowledge about the remaining life-expectancy of the tundish helps rescheduling the production.
Optimal scheduling is therefore only possible when the tundish's lifeexpectancy is known in advance. A detailed analysis of the scheduling problem can be found in (Dorn et al., 1993). Mathematical-analytical methods as used in Operations Research approaches are often insufficient for planning problems. This is due to three reasons: The imprecise information in the production process, combinatorial complexity of the search space, and conflicting objectives for production optimizing. The combination of several knowledge-based techniques, especially approximate reasoning and constraint relaxation, is a promising way to handle these problems.
In (Slany et al., 1992) we use a case study to demonstrate how knowledgebased scheduling works with the desired capabilities to schedule short-term production. The applied knowledge representation technique covers the vagueness inherent in the problem domain by using fuzzy set theory. Based