Abstract
This research focuses on the problems of Simultaneous Scheduling of Machines and AGVs in FMS. According to the problem, different system characteristics will affect the performance of system. In FMS, there are most multi-functions CNC machines, so same process in one work can choose the alternative machines to do. This characteristic can increase the flexibility of scheduling more. Alternative machines affect the makespan and the using percentage of machines and AGVs. In past researches, when using Mathematical Programming to solve simultaneous scheduling problem, transportation time of AGVs is the transportation distance divided by the speed of the AGV. It didn’t consider the actual problems that AGVs will delay or be in Deadlock due to the congestion during the transportation. Therefore, for the problems of Simultaneous scheduling of Machines and AGVs in FMS, this research constructs the discrete event simulation to achieve actual conditions, and use zone-control to deal with the problems that lead to deadlock. If the research considers more system characteristics, the fidelities of system will increase too. So, when considers the condition of alternative machines and zone control, there will be two low-fidelity models and one high-fidelity model. This research will cite the Multi-fidelity Optimization with Ordinal Transformation and Optimal Sampling (MO2TOS) optimization framework. It will effectively use the relationship between low-fidelity models and high-fidelity model to solve the problems of simultaneous machines and vehicles scheduling in FMS. There are four sampling methods in MO2TOS framework (Greedy、Random、Hybrid、Adaptive Sampling). Because there are some relationships between models but not completely unrelated, using Greedy Sampling may consume less computing cost in high-fidelity model and have higher success rate to choose the best design than other methods. Greedy Sampling can also apply in other different FMS environments. In the MO2TOS framework, if great and bad designs are included in the same group, it may lead the mean performance of group to be worse. In the consequence, less sampling resources will be allocated into the group, a risk that can’t choose the best design may happen. So, this research further optimize the allocation methods of sampling resources in high-fidelity model. Use the performance of the best design in every group to be eigenvalue to conduct the sampling resources allocation will effectively allocate the sampling resources to critical groups and saving the sampling resources.