1. Minimisation of the system workload   unbalance.

2. Minimisation of the system unbalance and  the  number of late jobs.

For both objectives, heuristic methods were developed and the system performance was compared with exact mixed integer programming solutions. A simulation model using  GPSS  V was also developed to determine the system performance and influence of loading policies on dispatching rules employed to sequence the jobs. Only one of the proposed loading procedures was compared with the FCFS rule and another  one  was not used in the simulation model. The model consisted of four machines, and  parts entered into  the system in  batches of  5  to

15. Due dates of parts were assigned using the TWK policy. Performance measures were CPU time and machine utilisation. The dispatching rules used in the simulation model were FIFO, SPT, LPT, and MOPR. Shanker and Tzen  reported  that  the SPT rule performed the best on average. Although due dates were assigned to the incoming parts, no due-date-based dis- patching rules or measures of performance were used in the model. It is  unclear  why  the  authors  considered  due  dates for parts.

rules.

Ro and Kim [84] considered  the  FMS scheduling  problem as a process of two loading and four dispatching  subproblems. A hypothetical FMS was composed of four CNC machining centres, each of which had a finite buffer space, a load/unload station, and two AGVs. Two loading  subproblems were stated as follows:

1. Part type selection during initial entry (the EDD rule was used for the simulation experiments as a part type selection rule  during  initial entry).

2. Part type selection during general entry (the part having the highest ratio of remaining requirement to their original requirement would be selected   first).

Four dispatching subproblems  were stated as   follows:

1. Part-to-machine allocation rule (the SPT rule was used for the simulation experiments as a part-to-machine selection rule).

2. Process or machine centre selection rule (ARD, ARP, ARPD, NAR, and WINQ rules were used as process selec- tion rule).

3. AGV dispatching  rule  (a  heuristic  rule  was  used to select a  free AGV).

4. AGV route selection rule (a bi-directional guide path was used).

Ro and Kim proposed three process selection rules for evalu- ation. The performance of these rules was then compared with WINQ and NAR rules. The developed rules assumed that a number of alternative machines were available when an unex- pected event, such as machine breakdown, ccurred and thus, tried to take advantage of the inherent routeing flexibility of FMSs. The performance of developed rules was evaluated by building a simulation model of the FMS using SLAM II and FORTRAN. The model consisted of four CNC machines with a finite buffer space for each one. The simulation results indicated that, first, the ARD led to the best result in four performance measures except for system utilisation. Secondly, ARD, ARPD, and WINQ produced significantly better results than ARP and NAR in every performance measure. The authors demonstrated the effectiveness of the proposed rules for mul- tiple performance measures including makespan, mean flow time, mean tardiness, maximum tardiness, and system utilis- ation.

Gupta et al. [85] explored the applicability of multi-criteria approaches to the production scheduling problems of an FMS, and reviewed the pertinent literature on scheduling of FMS involving multiple objectives, and   discussed: