A Similarity-Based Quasi-ART Neural Network Clustering Method and Its Application on Part Identification and Manufacturing Cell Classification

• Main Authors: HUANG, KUO HUA, 黃國華
• Other Authors: YANG, FENG CHENG
• Format: Others
• Language: zh-TW
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/61538872962480828347

碩士 === 國立臺灣大學 === 工業工程學研究所 === 87 === This thesis presents a similaroty-based clustering method and a complete process flow of part and machine grouping to form manufacturing cells. The proposed method is adapted from the ART neural network model for object clustering. The presented quasi-ART method can avoid the dimilishment of elements “1” of the backward weights specified in an ART model, which result in incorrect grouping. A clustering effectivness is established in this research. Based on the factor compulation of internal closeness and external dispersion, the clustering effectiveness metric provides an effectin\ve judgement of clustering results. The ART neural network clustering method and quasi-ART method can both ehance their clustering capability by automatically setting the vigilance, based on the judgement of the proposed effectiveness metric. A rectangle ecpansion technique is illustrated to extract single manufacturing features from a part of compound features, to identify the pocess types imposed on the part. The proposed process flow of part and mechine grouping starts from the process type identification to construct a process type-part relationship matrix. Applying the quasi-ART clustering method to divide the parts and process types, part-process type grouping blocks are generated. Combined with the machine-process type relationship, the completed part and machine grouping is resulted. This process flow focuses on the corelation between process types and parts in stead of the simple machine-part relation. Therefore, the formed manufacturing cells contain manufacturing information about part types, process types, machine types and machine allocation fruction.