模糊自適應共振類神經網路於工件分族及機器分群上之應用：演算法之修正及效益評估

An Application of Fuzzy ART Neural Network for the Part-Machine Grouping Problem ： modified algorithm and performance evaluation

郭仲倫(Chung-Lun Kuo)

指導教授 ： 葉若春 ； 鄭春生

元智大學/工程學院/工業工程與管理學系/碩士(1994年)

摘要

群組技術是實施彈性製造系統的成功關鍵之一。實行群組技術的主要利益
是簡化工廠內物料搬運流程。根據工件的相似性來分群，不僅減少了機器
設置的時間，也縮短了產品生產的等待時間和降低再製品。本研究在探討
模糊自適應共振類神經網路應用於群組技術之工件分族及機器分群問題。
此方法能處理二元值及連續型數值的機器/ 工件矩陣問題，並具有許多優
點，特別是此方法減少了複雜的計算和具有處理工業上大型分群問題的能
力。此方法的缺點主要在於進行分群時受到資料輸入順序的影響，特別是
在瓶頸機器出現時，可能使分群結果不佳。本文提出改良式模糊自適應共
振類神經網路演算法來增強原演算法處理分群問題的能力。本研究中採用
兩個評估指標來評估分群演算法所得的分群結果，他們是分群效益值及例
外元素的個數。

關鍵字

類神經網路 ； 群組技術

並列摘要

Group technology (GT) is one of the key issues in a
successful implementation of flexible manufacturing systems
(FMSs) . A major benefit of GT is the simplification of the
material flow within the shop.This fact coupled with reduced
set-up times , which result from part similarities , yields
shorter lead times and lower work-in-process . This study
investigates the application of Fuzzy ART neural network to
the part-machine grouping problem in GT . Fuzzy ART neural
network provides a framework for both binary and continuous
values , and offers several advantages , particularly the
reduction in computatioal complexity and ability to handle
large scale industrial problems . One weakness of this
approach is that the quality of a grouping solution is
mainly dependent on the initial disposition of part-machine
incidence matrix especially in the presence of bottleneck
machines . A modified Fuzzy ART neural network has been
developed to enhance the Fuzzy ART neural network in
part-machine grouping problem . In this study,there are two
measures used to evaluate the quality of solutions given by a
cell formation algorithm.They are grouping efficiency and the
number of exceptional elements.

並列關鍵字

Neural Networks ； Group technology