由於工業電腦產業為少量多樣的生產型態,使得SMT(Surface Mount Technology)生產系統在換線作業中面臨重大挑戰,時常未能在下一張訂單開始之前完成人工備料及綁料作業,導致昂貴機台閒置及產能利用率降低。因此本研究利用粒子群最佳化發展出一套BPSO(Binary Particle Swarm Optimization)clustering群集分析方法,其結合區域搜尋方法K-means,以加快找到全域最佳解,且加入輪盤選擇的操作,減少陷入區域最佳解的現象。 本研究分別先採用4個已知群集分佈的基準資料集Iris、Wine、Vowel及Glass作為比較依據,驗證不同群集方法分群的準確性,再將其應用在研華科技新店單板廠機種BOM(Bill of Materials)表資料,藉由產品族的概念,找出相同族群的機種共用料,將同一族群之機種安排在一起生產,便能縮短SMT換線作業時間,以因應市場少量多樣的生產模式。
Due to characteristics of high mix low volume for industrial personal computer (IPC), SMT (Surface Mount Technology) production system faces a great challenge that the engineer can not complete the material preparation process before next order coming. It results in idling expensive machine and decreasing capacity utilization. Therefore, this research intends to employ particle swarm optimization (PSO) with integrating K-means method and roulette selection operator in order to accelerate searching speed and avoid falling into local solution. For the purpose of evaluation, we first use four benchmark datasets- Iris, Wine, Vowel, and Glass as the comparison basis in order to verify the accuracy for each clustering method. The simulated results have shown the proposed method is able to cluster the data more precisely than the convention methods, like PSO and K-means. Moreover, the proposed method is applied to cluster the BOM (Bill of Materials) for Advantech Company which is a very famous IPC manufacturer. According to the concept of product family, the clustering results can be used to find the shared materials for each product family. Thus, the engineer can arrange the orders belonging to the same family to be manufactured together in order to reduce the SMT setup time. The results have shown this capability for the proposed method.