| Summary: | 碩士 === 國立臺灣大學 === 工業工程學研究所 === 89 === Because less contamination, shorter cycle times, and the flexibility of changeover, cluster tools play an important role in modern fabs. How to evaluate the capacity of cluster tools accurately and how to improve their throughput efficiently are two key issues of keeping competitiveness in the fab industry.
This thesis considers cluster tools as the extension of the traditional job-shop scheduling problem. Based on active schedule generation algorithm proposed by Giffler and Thompson at 1960, this thesis integrates blocking constraints of cluster tools and the backtracking technique to develop a schedule enumeration algorithm that can generate all feasible non-delay schedules for a single-gripper cluster tool when the lot size is small. For realistic problems with a standard lot size (22 to 25), two scheduling heuristics are developed.
The first is called schedule extrapolation algorithm. This thesis proposes a criterion to identify periodical operations of a schedule of a cluster tool. First, find a periodical schedule with a small lot size that is generated from schedule enumeration algorithm above. Second, construct a complete schedule by repeating periodical operations.
The second is called FAFS-LA + MIF algorithm. FAFS-LA and MIF are two constraints, and stand for First Available First Serve-Look Ahead and Moving-In First respectively. This algorithm embeds these two constraints in schedule enumeration algorithm.
In this thesis, makespan and total flow time are chosen as the objective functions. According to empirical results, these two heuristics have the same performance for makespan, but FAFS-LA + MIF is better for total flow time.
|
|---|
