A KNOWLEDGE-BASED APPROACH TO DYNAMIC JOB-SHOP SCHEDULING

Job-shop scheduling has challenged many researchers in that, although it belongs to a class of problems amenable to combinatorial analysis, a fully mathematical solution would not be feasible for many applications of realistic size and scope. A plethora of approaches to computer-assisted job-shop scheduling have been proposed, ranging from OR-type optimization techniques to AI/knowledge-based satisficing solutions. However, few of these approaches have had any success in delivering generic operational solutions. Logica, as a part of a three-year ESPRIT project (No. 418) in open CAM systems (OCS), has developed a software tool for dynamic finite capacity job-shop scheduling (ESPRIT 1989). The approach adopted in this development combines conventional schedule generation with knowledge-based evaluation and repair, to provide an integrated decision-support tool for the managers of job-shop environments. The system can interpret disturbance information acquired from on-line shop monitoring computers, and can suggest repairs to the current schedule in order to minimize operational disruptions. In addition, the system provides automatic schedule evaluation and improvement facilities, using expert knowledge, and flexible graphical editing for manual schedule modification. The work included a systematic analysis of the generic types and structures of knowledge involved in schedule evaluation and repair. It produced a set of guidelines for structured knowledge elicitation in similar application domains. Development involved integration of diverse software techniques (logic programming, relational databases, object-oriented programming, algorithmic programming) and tools [Prolog, Oracle, NeWS, ‘C’). This resulted in a powerful prototyping environment for knowledge-based or conventional production scheduling applications. The range of scenarios used during the design and development of the software included one based on an automated cell with on-line computerized production monitoring systems. This was linked to output from an MRPII system, provided by a large international manufacturer of printing machinery based in the UK, who also participated in the evaluation of the results produced. The results show that this hybrid approach to production scheduling is a promising basis for developing practical decision-support tools for managers of job-shops. This is because the approach can be easily configured to use various optimization algorithms or heuristics, providing performance advantages. together with the flexibility of knowledge-based techniques which facilitate multiple-criteria scheduling and using knowledge about ‘soft’ constraints for schedule refinemerit. © Taylor & Francis Group, LLC.