應用關聯規則探索互聯工廠製程失效模式與影響分析之研究

Failure Modes and Effects Analysis through Association Rule Mining in Hyperconnected Manufacturing

謝承桓(Cheng-Huan Hsieh)

指導教授 ： 林小甘

義守大學/管理學院/工業管理學系/碩士(2018年)

摘要

互聯工廠實現了與客戶相連，目標是從“產銷分離”到“產銷合一”，滿足客戶無縫化、透明化、可視化的最佳體驗。主動權由生產者轉向客戶，從大規模製造轉變為大規模訂製，從企業為中心轉為以客戶為中心，創造有效需求，有效供給。如何提升互聯工廠製造品質與產品良率， 縮短產品上市時程變得格外的重要。故互聯工廠視品質為核心競爭能力之一，因此運用失效模式與影響分析 (FMEA)於製程的改善是絕對必要的。藉由實施 FMEA將公司日積月累的資料透過分析來發掘產品的異常關聯規則，提供給公司一個具系統化、科學化與量化的參考資訊，公司管理者能根據特定的產品以及異常狀況來推測是哪個環節出了問題，同時提出改善對策以防止故障再度發生，進而提升產品之可靠度。因此本研究將某半導體封裝與測試製造服務公司為研究驗證案例，此公司提供包括半導體晶片前段測試及晶圓針測至後段之封裝、材料及成品測試的一元化服務，主要產品有：AS3、DOFU、Module、FCCSP、LBGA、PBGA、SD、WBGA，本研究將此8項生產線資料，套入R語言中的arules套件進行Apriori演算法來進行分析，希望從資料庫中挖掘出有價值的資料，尋找出各產品異常的主要狀況以及各異常狀況是否有關聯規則，此8項產品主要的異常狀況及關聯規則，已詳述於研究結果章節，但在失效模式與影響分析中，風險級數以及難檢度主要是由專家知識或經驗來取決，因此後續的研究希望藉由本研究結果及專家知識智能模型(Intelligent Model)，更快的算出風險優先級數，做出完整的失效模式與影響分析。

關鍵字

互聯工廠 ； 失效模式與影響分析 ； 關聯規則 ； R語言

並列摘要

Hyperconnected Manufacturing has achieved connecting with customers, the goal is to transform “Production and sales separation” into “Production and marketing integration”, it can satisfy customer’s best experience from seamless, transparency and visualization. Initiative shifts from producer to customer, from large-scale manufacturing to mass customization, from enterprise-centric to customer-centric, create effective demand, effective supply. How to improve the manufacturing quality, product yield and shorten the time-to-market of the product becomes extraordinarily important to the Hyperconnected Manufacturing. Therefore, the Hyperconnected Manufacturing regards quality as one of the core competitiveness, so it is absolutely necessary to use Failure Modes and Effects Analysis (FMEA) to improve the process.
Through implementing FMEA, the company's accumulated data is used to analyze the abnormal condition association rules of the products, and provide the company with a systematic, scientific and quantitative reference information. The company's managers are able to surmise which part of the problem is caused by specific products and abnormal conditions, and propose improvement measures to prevent the failure from happening again, thereby improving the reliability of the product. Therefore, this study will be a semiconductor packaging and test manufacturing company for research and verification cases, the company provides a unified and customized service including semiconductor wafer front-end testing and wafer needle testing to the final stage of packaging, materials and finished product testing. The main products are: AS3, DOFU, Module, FCCSP, LBGA, PBGA, SD, WBGA, in this study, the eight production line data were put into the arules suite in R for Apriori algorithm for analysis. I hope to dig out valuable data from the database, find out the main conditions of each product anomaly and whether there are related rules for each abnormal situation. The main abnormal conditions and association rules of these 8 products are related to the research results section. However, in the FMEA, the risk level and the difficulty level are mainly determined by expert knowledge or experience. Therefore, the subsequent research hopes that the risk priority can be calculated faster and the complete FMEA can be made by the results of this study and the Intelligent Model.

並列關鍵字

Hyperconnected Manufacturing ； Failure Modes and Effects Analysis (FMEA) ； Association Rule ； R

參考文獻

英文部分

Agrawal, R. and R. Srikant (1994). "AFast Algorithms for Mining Association Rules." Paper presented at the Proceedings of the 20th VLDB Conference on Santiagode Chile, Chile.

Carlson, W. D., et al. (2001). "Potential Failure Mode and Effects Analysis." AIAG.

Feigenbaum, A. V. (1983). Total quality control, New York,NY:McGraw-Hill.

Han, J., et al. ( 2000). "Mining Frequent Patterns without Candidate Generation." Proceedings of 2000 ACM-SIGMOD International Conference Management of Data 29(2): 1-12.