Abstract
The TFT-array process is a portion of the TFT-LCD manufacturing. The scheduling problem of the TFT-array process can be treated as flexible job shop with recirculation. A flexible job shop is a generalization of the job shop. Each workstation consists of multiple parallel machines and the machines have similar functions. The machines in a workstation may have different efficiency. A job has to choose a qualified machine when entering a workstation. In addition, a job may visit some workstations more than once. To solve this difficult problem, this thesis uses two methods. One is called “traditional two-way heuristic”, which is proposed by Brandimarte【1993】. The other one is called “modified two-way heuristic”, which is proposed by this study. The meaning of “two-way” is that we separate the problem into two dimensions. One is machine allocation and the other one is job shop scheduling. During the solution process, the solution of the procedure of one dimension problem will become the initial solution of the other one. The procedures for these two dimension problems keep alternative in loops until satisfying the over all stopping criteria. We refer the method which is proposed by Brandimarte【1993】 as the traditional two-way heuristic. It uses the stochastic tabu search to solve the job shop dimension. In the machine allocation dimension, it chooses one operation on the critical path and this operation is assigned to one of the alternative machines. It chooses one operation at one time and records the new solution. When all operations on the critical path are assigned, the best insertion solution is chose. In modified two-way heuristic, we use the fast tabu search proposed by Nowicki and Smutnicki【1996】 to solve the job shop dimension and use the simulated annealing to solve the machine allocation dimension. The experiment results show that modified two-way heuristic outperforms traditional two-way heuristic when solving medium and big problem.