Montazeri and Van Wassenhove [32] stressed the need for simulation prior to setting up the FMS. The characteristics of a general-purpose, user-oriented discrete-event simulation for FMS were discussed. Following a brief review of the literature on scheduling rules, they analysed the performance of 14 dispatching rules using the developed simulation model  to mimic the operation of a real-life FMS.  They  studied  SIO, SPT, SRPT, LRPT, SMT, LDT, FRO, LIO, LPT, SDT, LMT,

MRO, FIFO, and FASFO scheduling rules in a  simulation model consisting of five machine centres and three loading/unloading stations, and one WIP buffer with 11 pos- itions. The criteria chosen for evaluating scheduling rules were average and variance of waiting time per part, average and variance of machine utilisation, average buffer utilisation, aver- age shuttle utilisation, average carrier utilisation, and makespan. Four decision points were used in the model including select next part to be processed by the machine, select next part to be moved in the system, select next part to be reclamped by the worker, and select next part to be  loaded  on  the carrier from the facility. All decision points were assigned the same priority rule in every run. The authors confirmed that a combi- nation of a time-based rule like  SMT  and  a  due-date-based rule like SLACK merits further  research.  However,  they did not conclude which rule is the  best  regarding  all  perform- ance measures.

Kim [62] studied a manufacturing system in which alterna- tive operations were considered. The author tested four different methods to define the precedence relationships. Dispatching rules considered in this study included SPT, Slack, S/RMOP, S/RMWK, MDD, MOD, COVERT, and ACT. Only three performance measures were used, i.e. mean flow-time, mean tardiness, and number of tardy jobs. The author concluded that the MDD rule outperformed all other rules. However, the results showed that the MDD rule did not perform well under one  specific  precedence method  mentioned above.

Hutchison and Khumavala [63] compared seven  real-time and two off-line scheduling schemes for a random FMS with a dynamic environment. Results indicated that full advantage could be taken of these flexible systems if  off-line schemes were applied. The study was conducted  in  two  phases.  The first phase compared the two off-line scheduling schemes using 100 randomly generated static test problems to determine the deterioration in performance that would result from decompos- ing the problem. The second phase compared the decomposed off-line scheme against the seven real-time scheduling schemes using simulation. The scheduling rules for the second phase included SPT, LWKR, MOPR, SPT/TOT, MWKR, LOPR,   and

LPT. Only two performance measures were employed, namely, average flow-time, and adjusted production rate that was similar to makespan. They concluded that SPT was the best real-time scheduling rule for the adjusted production rate, and the MWKR rule was the best rule for the average flow-time. However, only processing-number and  time-oriented-based rules and performance measures were used and no due-date information was  employed in  the model.

Ishii and Talavage [64] proposed a transient-based algorithm for determining the length of the simulation window. This algorithm defined a short-term scheduling  interval  based  on the system’s transient state that was evaluated by a measure similar to the workload in the  FMS.  As  opposed  to  the Wu and Wysk [65] approach, which  is  explained  in  Section 4.3, the size of the scheduling interval was considered as a  variable in the Ishii and Talavage [64] approach. The achievements are similar to those of Wu and Wysk [65] work.

Hutchison et al. [66] strongly advocated the use of off-line scheduling rather than real-time scheduling. The influences of scheduling schemes and the degree of routeing flexibility on random job shop FMS in a static environment were investi- gated. Three schemes were tested, which include two off-line and one real-time under different routeing flexibility. In  the third scheme, they used a dispatching rule, SPT, coupled with a look-ahead policy to establish a schedule  in  a  real-time mode. The SPT rule was used  as  a  queue  dispatching  rule, and also to determine when parts were released into the system. The only performance measure was makespan. The results showed that the real-time scheme had an average makespan of 34.40% larger than the first off-line scheme, and 30.83% larger then the second off-line scheme. These differences became larger when the routeing flexibility increased. They did not mention why they did not use more performance measures or why they did not test off-line schemes with more dispatching rules.