were dispatched within the system by using the FIFO rule. Performance measures used in the simulation model included production rate, flow-time, makespan, and machine utilisation. The system was modelled on a Sun workstation using a discrete-event simulation language written in LISP. The fuzzy rules were also written in LISP as a subroutine  and  called within the simulation model. An example of fuzzy rule  used was  stated  as follows:

If “the utilisation ratio is less than low”, and “the part- processing ratio is more  than  moderately  high”  then  “force the part”.

Unfortunately, owing to the confidentiality of the work, the authors did not disclose details. It was concluded that the look- ahead procedure led to better results  than  fuzzy  rules. They did not mention whether they had tested different fuzzy rules or fuzzy functions, used in the rule base, for further improvement. Kova`cs et al. [102] described the application of expert sys- tems to assist quality control and to help the control of FMSs via simulation. They believed that a close-to-optimal operation of complex, real-time, and stochastic systems such as FMSs could not be achieved by the application of traditional program- ming. They strongly advocated the use of expert system and artificial intelligence techniques in conjunction with sophisti- cated modelling and simulation. Their simulation model con- sisted of four machines, two robots, one input store, one output store, and one AGV to manufacture  four  different  part  types of  workpieces.  SIMAN/CINEMA  was  used  as  a   simulation

package  with the  advice  of  expert systems.

Baid and Nagarur [92] strongly advocated the use of simul- ation techniques and described an intelligent simulation model- ling for a manufacturing system, considering a three-level interdependent planning hierarchy. The developed intelligent simulation system incorporated three  basic  modules, namely, an intelligent front end, a simulator, and  an  intelligent back end. The intelligent front-end module consisted of a simulation model generator. The model generator created a SIMAN model and experimental framework for the simulation of manufactur- ing systems. An intelligent back end was employed to analyse the output of the simulation. The simulation model reported in this study contained four workstations, one automated storage/retrieval system (AS/RS), one vehicle in AR/RS, AGVs for handling of parts, and one loading/unloading station. One user interactive rule was provided which gave a choice to the user, to order the jobs personally using some chosen criteria such as due-dates, if necessary. Three traditional dispatching rules, FIFO, LIFO, SLACK, were modelled. The intelligent decision support system developed by Baid and Nagarur was not  able  to  interpret  the  results,  and  also  failed  to      select a combination of scheduling rules as the best one for all performance measures.

Wang et al. [103] presented a system-attribute-oriented knowledge-based scheduling system (SAOSS). The SAOSS employed an inductive learning method to  induce  decision rules for scheduling by converting corresponding decision trees into hidden layers of a self-generated  neural  network.  The FMS under consideration was composed of two machines, a machining centre, five tools, four fixtures, and two industrial robots for loading/unloading parts. Six part dispatching rules,

which  were SPT, LRPT, SRRTIOM, EDD,  MRTRAUO,    and

FCFS, were compared with the proposed SAOSS. Performance measures employed were mean flow-time and mean tardiness. Results indicated that the SAOSS was superior to other dis- patching rules in both performance   measures.