Min et al. [109] developed an FMS scheduler, which  applied a neural network to present fast but good decision rules, to maximise the desired values of the objectives. The scheduler generated the next decision rules, which were based on the current decision rules, system status, and performance meas- ures. The FMS consisted of four machine centres, a washing machine, 39 work-in-process (WIP) storage racks, and a crane for material handling. Nine performance criteria were con- sidered, namely, mean tardiness, maximum tardiness, mean flow-time, average machine utilisation, average crane utilis- ation, average total processing time, slack, average jobs in the system, and average WIP in the rack. The model was developed using SLAM II. Results showed the comparison of mean tardiness, maximum tardiness, mean flow-time, and slack between the values obtained from the proposed neural network, and the values obtained from selection of next decision rules randomly. However, Min et al. concluded that the methodology had difficulty to achieve all the objectives simultaneously. Kim et al. [110] employed the same configuration to study an integrated approach of inductive learning and neural networks for developing a multi-objective FMS scheduler.  Results showed that the proposed approach gave better results than the neural network approach that was developed by Min et al. [109].

Chen et al. [111] presented an intelligent manufacturing scheduling and control with specific applications to the load/unload operation of an AGV system in a real FMS. The FMS had two enter/exit areas, four load/unload stations, one storage rack for WIP inventory, and two AGVs. A neural network provided the material handling control strategy. Data obtained by simulating various scenarios were used to train the artificial neural network. The trained neural network generated appropriate output for a particular input. The control strategy was  simulated  and  compared  with  a  static  system  that using

LOPNR rule for load/unload stations, and SPT for AGV. The performance measures employed were flow-time, throughput, time in load/unload station, time wait for load/unload station, WIP rack, and AGV queue size. Results showed that the proposed control  system  was superior  to the static  system  as it led to shorter flow-time, higher system throughput, and less WIP inventory.

Yu et al. [112] proposed a fuzzy inference-based scheduling decision approach for FMS with multiple objectives, which consisted of different and dynamic preference levels. The pref- erence levels were dynamic because the priority given to different objectives might change depending  on the conditions of the production environment, such as an abnormally large number of customer orders. The changes in production environ- ment were sensed by environmental variables and these changes were input in a fuzzy inference mechanism, which output the current preference levels of all objectives. A multiple criteria scheduling decision was then made, using the partitioned com- bination of the preference levels. A simulation example was used to demonstrate the proposed approach. The FMS consisted of three machines and five different products with routeing flexibility. Two objectives were considered, i.e. mean flow- time, and absolute slack, the latter was used to penalise both tardiness and earliness in a just-in-time  system.  The  system was simulated in the “C” language. The proposed fuzzy inference-based scheduling rule was compared with two tra- ditional dispatching rules, which were earliest finishing time (EFT) and shortest absolute slack (SAS). Results indicated that the proposed fuzzy rule produced the best result for all perform- ance measures, except mean flow-time for a light workload situation. It was concluded that the proposed fuzzy rule had a very robust  performance  under a  heavy workload.

Qi et al. [113] described the use of parallel multi-population GAs to deal with the dynamic nature of job-shop scheduling. A modified genetic technique was adopted by using a specially formulated genetic operator to provide an efficient optimisation search. The proposed algorithm was programmed using MAT- LAB. Four performance measures, number of tardy jobs, total tardiness, mean flow-time, and makespan were monitored for comparison. The proposed algorithm was compared with five conventional scheduling rules, which were EDD, FCFS, LSF, SPT, and LWR. Results indicated that the performance meas- ures were improved by using the proposed GA. However, the system configuration under evaluation was not mentioned clearly.