Scheduling problems have been and still are dealt with extensively in Operations Research. They abstract from most real world requirements thus delivering a mathematically concise description. Many different models have been developed and solutions been proposed. Most of them are aimed at finding a somewhat optimal solution regarding the overall processing time but will not normally scale up to solving real world problems.
The knowledge-based system WIZARD is designed to help human planners in creating schedules in real world production environments. It incorporates a great variety of requirements which are common in industrial applications but normally not dealt with in most OR-models. Multiple resources to any operation are considered, the time model incorporates elaborate shift patterns, jobs with and without priorities, release and due dates can be handled, and constraints can be assigned a weight thus representing hard requirements and soft preferences.
In addition to applying WIZARD to the real world problems it was designed for, its performance on some OR-problem instances was measured. The results show that larger problems can be solved in a rather satisfying way whereas the solutions to typical ``toy-problems'' are far worse than the optimal solutions.
In a three-year research project supported by the German Ministry of Education, Science, Research and Technology under contract 01IN511 WIZARD will be enhanced and evaluated in industrial environments. The results show that the representation formalisms are sufficient but that the knowledge used and the propose-and-exchange-algorithm must be improved to find better solutions than are found at present.