本研究主要提出一種新的群集分析方法-PSKO演算法,透過IRIS、Glass、Vowel 以及Image Segmentation等基準資料集與K-Means、PSO clustering、Hybrid PSO演算法比較群集效益。PSKO為本研究評估出最佳的群集分析方法。 在個案研究中,本研究以研華科技新店單板廠為例。由於少量多樣的產品需求型態,使得SMT生產系統在換線時,備料作業時常未完成,導致昂貴的機台發生閒置的情形。因此本研究應用群集分析方法,將所有機種的BOM表根據其特徵進行兩階段分群,在第一階段中先以自適應共振理論2神經網路(ART2)自動找出群數,接著第二階段中再以K-Means、PSO clustering、Hybrid PSO、PSKO各演算法找出最佳的分群結果,最後ART2+PSKO為最佳兩階段分群法。得知分群結果後,生管工程師將同一群集的機種排在一起生產,便能簡化備料作業,達到縮短SMT的換線時間之目的。
This study intends to propose a novel clustering analysis technique, particle swarm K-means optimization (PSKO), which integrates both the particle swarm optimization and K-means method. In order to evaluate its computational performance, three clustering analysis methods including K-means, PSO clustering and Hybrid PSO method are employed for comparison via IRIS, Glass, Vowel and Image Segmentation benchmark data sets. The simulation results indicate that PSKO outperforms these three methods both in speed and accuracy. For further assessing PSKO’s capability, a world-class industrial computer manufacturer, Advantech company, which belongs to the high mix low volume production system, provides the related evaluation information. Its production characteristic is that the material preparation process often has not completed during setup in SMT system. This results in expensive machine idleness. Thus, we apply a two-stage method, which first uses the adaptive resonance theory 2(ART2) network to determine the number of clusters and then employs K-Means, PSO clustering, Hybrid PSO, PSKO algorithms to find the final solution. The results show that the best two-stage method is ART2+PSKO. Through order clustering, the production planners can manufacture products in the same cluster in order to save the material preparation time and also achieve reducing SMT setup time.