反應六標準差水準之多品質特性同時最佳化演算法

Abstract

田口方法(Taguchi method)在提升產品品質的成效方面頗具實用價值，因此深受國內外產業界的肯定。但隨著產品的設計愈來愈複雜，產品品質的良莠問題已非最佳化單一品質特性就能解決，往往需要考慮到多品質特性的同時最佳化，而田口方法僅能解決單一品質特性最佳化的問題，因此陸續有中外文獻提出許多解決多品質特性同時最佳化的方法，但是這些方法都只著重於搜尋最佳之因子水準組合，並未能考量製程產品產出的良率在經過長期生產後，是否仍符合原訂目標值的問題。目前雖然已有文獻利用迴歸模式來解決此問題，但此迴歸模式並未考慮到雜音因子，因此，為了能使多品質特性製程最佳化，且同時達到六標準差的品質水準，本研究發展出一套能反應六標準差水準的多品質特性同時最佳化演算法，以找出在長期生產後，能符合六標準差水準之多品質特性最佳因子水準組合。本研究所發展之演算法是同時應用控制因子與雜音因子來建構各品質特性的反應模式(response model)，以估計製程的平均值與變異數，然後再利用以製程良率為指標的望想函數(desirability function)來達到最能符合六標準差水準之多品質特性同時最佳化的目標。本研究最後以一個密封元件產品之油封(Oil-Seal )製程最佳化的實例，驗證了本研究方法確實有效可行。

Taguchi method has been extensively used for enhancing product quality in industry. More than one quality characteristic must be simultaneously considered to effectively improve the product quality due to the product design is increasingly becoming complicated. However, Taguchi’s applications optimize only single quality characteristic. When more than one characteristic is to be optimized, engineers usually determine the optimal factor-level combination from their experiences. Although several procedures for optimizing multi-response process have been developed in recent years, these methods do not consider the deviation of process yield from the desired target caused by long-term manufacturing process. Therefore, this study proposes an optimizing procedure to resolve multi-response problem considering the deviation of process yield from long-term manufacturing process for achieving the six-sigma quality level. Initially, the effects of control factors and noise factors are simultaneously considered to construct a response model to estimate the process mean and the process variance for each quality characteristic. The desirability function based on process yield is then employed to determine the overall quality performance for multiple responses and thereby determine the optimal factor-level combination to achieve the six-sigma quality level. An experiment regarding to the development of oil-seal demonstrates the effectiveness of the proposed procedure.