An evolutionary simulation-optimization approach for the problem of order allocation with flexible splitting rule in semiconductor assembly

• Published in: Applied intelligence (Dordrecht, Netherlands) Vol. 53; no. 3; pp. 2593 - 2615
• Main Authors: Chiu, Chun-Chih, Lai, Chyh-Ming, Chen, Chien-Ming
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2023; Springer Nature B.V
• Subjects: Artificial Intelligence; Assembly; Computer Science; Computing costs; Decision making; Evolutionary algorithms; Genetic algorithms; Machines; Manufacturing; Mechanical Engineering; Operational problems; Optimization; Optimization algorithms; Performance evaluation; Processes; Search algorithms; Simulation; Simulation models; Splitting; Statistical methods; Semiconductor assembly; Order allocation and lot splitting size; Optimal replication allocation strategy (ORAS); Enhanced genetic algorithm; Simulation optimization
• ISSN: 0924-669X; 1573-7497
• DOI: 10.1007/s10489-022-03701-2

In this study, we propose an evolutionary simulation-optimization algorithm for the order allocation problem with flexible splitting rule in semiconductor assembly (SA). There are numerous complex production constraints associated with the operational problems in SA, such as identical and unrelated machines, flexible order lot split, and stochastic processing time, which hinder the decision-making process (i.e., which order allocates which machines and the most efficient lot-split size for a production system). To address complex production constraints in SA, this study constructed a simulation model to evaluate the system performance of each design alternative with minimization of the expected flow time of all orders. Due to the large design alternatives, this study proposes a simulation optimization algorithm to efficiently determine the design alternative. Owing to the high time consumption involved in using a high-fidelity simulation model to evaluate system performance, this algorithm employed a ranking and selection method known as the optimal replication allocation strategy (ORAS), to efficiently allocate computing resources. The ORAS reduced the additional computing cost of non-critical solutions and generated an elite set, which contained elite members not significantly different compared to the global best (Gbest), in each generation of the search algorithm. As this problem is complex and involves numerous local and global optima, an enhanced genetic algorithm (EGA) is proposed to utilize the elite set to enhance the diversity and further improve the solution quality. The proposed algorithm was validated by comparing its performance using statistical methods on 12 instances with those of several state-of-the-art algorithms. The results demonstrated the superior solution quality and search efficiency of the proposed algorithm compared to those of the competitors.