應用資料探勘技術改善PI製程之皺摺問題

Abstract

本研究利用資料探勘技術改善製程良率問題，研究對象為台灣某Polyimide(PI)薄膜製造公司。由於公司的產品PI薄膜上有嚴重皺褶的問題，造成高退貨率，因此本研究蒐集PI薄膜之製程參數及其最終產品之品質參數，利用決策樹方法—C4.5及1R挖掘出影響品質參數之關鍵製程參數並給予每一關鍵製程合理區間範圍，提供改善PI薄膜皺褶問題，此目標也是傳統用於良率提升之統計方法較難達成的部份。同時建構完整於製程良率提升之架構，除了包含資料探勘部分外，進一步說明如何應用資料探勘結果，實際進行良率改善之後續行動。 研究結果發現，利用原有之參數進行分析，縮小分析製程範圍，並進一步找到被工程師忽略而未加入分析之重要參數，進而將良率由52%提升至99%，幾乎完全消除PI薄膜的皺褶問題

This study applied data mining techniques with an empirical study in a polyimide film (PI) manufacturing company in Taiwan. In the manufacturing process, severe wrinkle problems on the PI film would result in unacceptable outputs. We collect the process and quality parameters data and use C4.5 and 1R in order to identify critical process parameters as well as their appropriate range for reducing the wrinkle problems. The data mining results indicate that two process parameters are critical to the wrinkle problems; however, their value adjustment may affect another quality attribute. By analyzing the manufacturing operation associated with the two critical parameters, we propose the inclusion of a new manufacturing operation. With this inclusion, we can significantly reduce the wrinkle problems, with yield rate improved from 52% to 99%.