當前許多製造業者為了維持與顧客間的良好關係,必須快速處理客訴問題,而製造業者主要利用改正行動處理單 (corrective action request, CAR),進行客訴問題的資訊蒐集與分析。當外部顧客及內部顧客提出改正行動處理單時,製程工程人員憑藉著過去的經驗或者對相關製程及產品的知識,提出相關品質問題之解決分析報告。 本研究主要針對製造業提出一個品質診斷系統,並且以印刷電路板業為例,進行研究。本研究所提出之品質診斷系統,主要是以資料挖掘技術為基礎,診斷出各式各樣來自於顧客所提出之產品缺陷原因。以外部改正行動處理單來說,記錄著對於所有產品缺陷的描述及改正措施,研究中,運用自組織映射網路,針對外部客訴問題做工作站之聚類。對於分析主因方面,本研究藉由決策樹分析之運用,找出產品異常主因。以內部改正行動處理單而言,本研究蒐集各工作站之站內資料,並運用決策樹分析,產生出品質問題之相關診斷法則。本研究實際以一PCB製造商為案例研究對象,來驗證本研究所提出之架構,對於製程工程人員分析品質問題時,是一有效之運用工具。 關鍵詞:資料挖掘、改正行動處理單、自組織映射網路、決策樹分析、品質診斷系統、PCB
A quick response to corrective action request (CAR) from customers is an important issue in today’s manufacturing in order to maintain a good customer relationship. When internal or external customers issue corrective action requests, the process engineers must prepare a report using past experiences or related process/product knowledge. The focus of this research is on the development of a quality diagnosis system for the printed circuit board (PCB) industries. A quality diagnosis system based on data mining technology will be developed to address various types of defects described by customers. External CARs that record the descriptions of defects and correction procedures will be collected and clustered by workstations using SOM neural networks. A decision tree will be applied to build a diagnosis knowledge base to address the root causes of defective products. For internal CARs, data are collected for each workstation and diagnosis rules will be developed using decision tree approach. Data from a local PCB manufacturer demonstrate that the proposed approach is a useful tool in preparing a CAR report. Keywords: data mining, corrective action request, SOM, decision tree, quality diagnosis system, PCB