| Determining the optimal factor settings for continuous quality improvement often faces with many challenges. Robust parameter design has been well accepted as a process optimization technique. Although tremendous progress has been made, there still remains a significant gap to be narrowed. In particular, there was a controversy over the approach for adopting Taguchi philosophy; therefore, in this dissertation, the response surface methodology is incorporated as a base. Within this framework, the proposed schemes are viewed from the customer's perspective combining the truncated statistics since manufacturing processes generally involve that quality characteristics are required to meet the specifications. Studies illustrated the S-type characteristic with an upper specification limit, N-type characteristic with lower and upper specification limits, L-type characteristic with a lower specification limit and dual quality characteristics with a specification region which is intersected by two sets of lower and upper specification limits. Along this line, Weibull and skew normal are applied to describe manufacturing processes which are positively and negatively skewed for S- and L-type characteristics, respectively. In addition, most studies focused on multiple independent characteristics. In contrast, a bivariate distribution and a correlation component are added into the quality-loss-function-based scheme so that dual quality characteristics will not be treated as two separated univariate. The benefit coming with the loss-function-based scheme is that it might avoid result in suboptimal solutions by allowing the deviation far from the target rather than forcing to zero bias. However, on some occasions, experimental region is subject to constraints. Accordingly, twenty different optimal designs were evaluated and compared in the study so that it can be an alterative whenever the experimental region is irregular. |
