based system outperformed the  priority  rules.

Watanabe [98] developed a new algorithm using fuzzy logic to determine the priority of parts in computer integrated manu- facturing systems for producing many kinds of products  in small volumes. The author believed that some customers require short due dates and do not mind paying high costs, whereas some do not mind the lateness of due dates. In this connection, the quickness (how fast a part should be produced) requirements and the profit of production indices were   studied.

In addition, priority of a part  in  a queue  was  determined by two  fuzzy rules:

1. If “the slack time is short” and “the profit index is  high”, then “the priority of the job is high”.

2. If “the slack time is long” and “the  profit  index  is low”, then “the priority of the job is low”.

The fuzzy propositions “slack time is short or high”, “profit index is high or low”, and “priority is high or low”,  were defined by membership functions. The results of the constructed simulation model showed that profit was significantly improved by using the proposed method in comparison with the SLACK rule, flow-time was increased by 20% but no other performance measure was used. Watanabe  did  not  compare  the efficiency of the proposed method with some other  dispatching  rules, such as SPT, and no transportation and alternative routeings were considered.

Chandra and Talavage [99] proposed an intelligent dis- patching rule called EXPERT using an expert system. They conducted a simulation  model  using  the  RUNSHOP program to demonstrate the effectiveness of the proposed rule. The system contained ten machines with one general queue. A part, upon completion of an operation, was not routed to a specific machine, but rather was sent to the general queue. Thus, a machine had a global option to select parts which, in  turn, might be processed on alternative machines. When a machine became idle and a loading decision was required, RUNSHOP called a subroutine, which updated all status variables required for intelligent reasoning to select a job for the idle machine. Selection of the job depended on the current  objective  that could be either maximising work progress rate  or minimising the number of tardy jobs. The  effectiveness  of  the proposed rule was compared with some traditional rules including SPT, EDD, LSPO, and LRS.  Performance  measures  were  number of completed jobs, number of tardy jobs, total work, average tardiness per completed job, and average shop lead time factor that was equal to the ratio of total flow-time to the total actual processing time expended on the job. Simulation results showed that the proposed method outperformed all dispatching rules listed above except when the shop lead time factor was con- sidered. The main shortcoming of the model is that only one decision  point  was investigated.

Nakasuka and Yoshida [100] proposed a new learning algor- ithm for acquiring sufficient knowledge, which enables the prediction of the best rule to be used under the  current line status. In this algorithm, a binary decision tree was automati- cally generated using empirical data obtained by iterative pro- duction line simulations, and it decided,  in  real  time, which rule to be used at decision points during the actual production operations. Simulation results of its application to the dis- patching problem were discussed with regard to its scheduling performance  and  learning capability.

O’Keefe and Rao [101] reported an investigation into part input sequencing methods for a flexible flow system. Two new dynamic methods were developed, i.e. look-ahead simulation, and a fuzzy heuristic rule base. These two new methods were then compared with three simple static  sequencing  rules and one dynamic rule. The system consisted of ten machines, nine load/unload  stations,  and  AGVs  to  transport  the  parts.  Parts