Ph.D.s' Dissertations A Study on the Manufacturing System State Based Scheduler using Neural Network and Simulator
2015.07.29 12:28
| year | 1999 |
|---|---|
| author | Kitae Kim |
| Keyword | Scheduling, Simulator, Clustering, Knowledge Acquisition, Dispatching Rule |
| Abstract | The domestic manufacturing world has come under more and more pressure of the foreign and domestic many customers. Customers demand shorter lead times and a higher product variety without making concessions on product price and quality. To remain competitive, a manufacturing system needs to react adequately to perturbation of its environment and uncertainties in the manufacturing process. The control strategy of manufacturing system has a lot of influence on the efficiency of the manufacturing system. Specially the effective scheduling is the important control strategy in terms that it makes the expensive device be operated effectively. There are so many researches about scheduling. Many researcher have interested in the decision making using the simulation and the dispatching rule for scheduling because they are able to apply the real time scheduling. This thesis is the study on the manufacturing system state based scheduler that can apply the real time schedule using the most proper dispatching rule to the current system state. The thesis the three research topics for the development of the manufacturing system state based scheduler : the development of the simulator for simulation the manufacturing system, the clustering method for the manufacturing system state, and the search method for the most compatible dispatching rule to the manufacturing system state. Finally, the simulation experiment results are given for comparing the proposed method and other scheduling method using the simulation techniques. There are so many computer program language for simulation. It is difficult to learn the special simulation language. Specially in case that the target system is not the system that the special simulation language are developed on, it is very difficult to model the target system. And the domain specific simulator can model and simulate the specific domain. Even if the domain specific simulator is developed on the event driven simulation concept, the general and concrete developing method is unknown. The thesis discusses a method for the development of specific domain simulator. This approach employs state transition graph describing state transition of part entity in the manufacturing system, and using ideas from the object-oriented approach. Using this approach, a part state driven scheduling simulator is developed. If using the proposed method, a domain specific simulator can be developed with less effort. Dynamic strategies have more potential than static strategies because they can react to changes in the system state. In this thesis, the system state based scheduler which suggest the most proper dispatching rule to current system state be proposed. The proposed scheduler scheme that consists of 2 phases: knowledge acquisition phase and schedule execution phase. The knowledge acquisition phase has two steps: System state pattern classification step and Scheduling rule base generation step. The knowledge acquisition phase provides two kinds of knowledge: System State Pattern Knowledge base (SPKB) and Scheduling Rule Base (SRB). System state pattern knowledge base is the knowledge for classification the current system state pattern, scheduling rule base is the knowledge to suggest the most proper dispatching rule to the classified current system state pattern. System state variables are used to each system state pattern classifier made by each ART II network as input data. In result, a state vector consisting of six symbols presents a system state. At the schedule execution phase, System State(SS) Finder classifies the current system state pattern corresponding the system state variables. Dispatching Rule(DR) selector selects a proper dispatching rule to the system state pattern at each decision point. Scheduler makes a schedule with the dispatching rule. The decision point, in this thesis, is defined as the time when a machine needs to select a job from the waiting jobs. A real time performance is considered to acquire scheduling rule base. During one state, various dispatching rules are applied. Once other state is classified and a dispatching rule is applied, the real time performance during the interval from the start time and the finish time of the state is measured. The performance is the dispatching rules performance on the state. The simulation experiment results are given for comparing the proposed method and other scheduling method : single dispatching rule and multipass simulation. The thesis proposed the system state based scheduler as the more efficient scheduling method. For the development of the scheduler, the general developing method for simulator, the clustering method of system states and the new performance measure for scheduling are proposed. The proposed simulator developing method is useful for developing the various domain specific simulator, and the method for the system clustering and schdeuling knowledge acquisition is useful to the real time scheduling. |
| c | PhD |
Download :