Scheduling Problems to Minimize Makespan and Total Completion Time in Flowshop Environment with Learning Effects

• Main Authors: Chung, Yu-Hsiang, 鐘愉翔
• Other Authors: Tong, Lee-Ing
• Format: Others
• Language: en_US
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/23920650868849125159

博士 === 國立交通大學 === 工業工程與管理學系 === 100 === In traditional scheduling problems, the processing time for a given job is assumed to be a fixed constant no matter the scheduling order of the job. However, it is noticeable that the job processing time declines as workers gain more experience. This phenomenon is called the “learning effect”. The learning effect is extensively studied in scheduling field recently, and it can be classified into two types: “the position-based learning” and “the sum-of-processing- time-based learning”. The two types of learning effect can be considered alone or simultaneously in a scheduling problem. The position-based learning is studied in this dissertation because of its model is the pure learning model in theory. In addition, most of the studies on the learning effect are focused only on single-machine setting. However, numerous real-world industrial problems belong to flowshop scheduling problems, and dealing with the flowshop scheduling problems is more complex than dealing with the single-machine problems. Most scheduling problems aim at determining an optimal sequence to minimize the objective function. The makespan and total completion time are the objective functions that are often studied. As a result, this dissertation discusses two flowshop scheduling problems with position-based learning effect. The learning effects are identical on all machines, and the purpose is to minimize the makespan in the first problem. The learning effects are distinct for different machines, and the purpose is to minimize the weighted sum of total completion time and makespan in the second problem. In this dissertation, the branch-and-bound algorithm is proposed to seek the optimal sequence for the small job-sized problem. Then the dominance properties and lower bounds are proposed to accelerate the procedure of the branch-and-bound algorithm. For the large job-sized problem, two well-known heuristic algorithms, simulated annealing and genetic algorithm are utilized to yield the near-optimal sequence. In the end, the simulated experiments are examined to assess the performance of the algorithms proposed in this dissertation. The computational results of the proposed problems reveal that the objective value calculated from the optimal sequence under the traditional environment is larger than the optimal objective value in the environment with learning considerations. It implies the influence of the learning effect is notable for the problems proposed in this dissertation. Furthermore, the efficiency of the branch-and-bound algorithm ascends as the learning effect enhances while seeking the optimal sequence. The proposed genetic algorithm has the best performance among all heuristic and meta-heuristic algorithms in terms of the accuracy. In addition, due to the large variance and the right skewness for the distribution of the execution time, the branch-and-bound algorithm is recommended to obtain the optimal sequence in a reasonable amount of time, or to derive the near-optimal sequence from the proposed genetic algorithm. Eventually, assigning the operator with stronger learning effect to the machine with heavier workload might derive smaller objective value while the operators are allocated in the flowshop environment.