Multivariate robust parameter design

In a product system, large numbers of design variables and responses are involved in performance analysis. Relationships between design variables and individual responses can be complex, and the outcomes are often competing. In addition, noise from manufacturing processes, environments, and misusage causes variation in performance. These issues must be considered before designs are finalized, and the design team must achieve the following objectives: (1) Meet performance target at nominal design, (2) Minimize variability of all vehicle attributes, (3) Produce the least incidences of "out-of-tolerance" conditions.; The proposed method utilizes the two-step optimization process from robust design and performs the optimization using Hotelling's T 2-statistic. Specifically, the proposed method achieves these objectives by: (1) Design and conduct simulation experiments using robust design methodology. (2) Transform the n responses from the experiments into a multivariate index T2O which measures the generalized distance of the sample from a target. (3) Decompose T2O into two (2) components, T2M and T2D , which represent the generalized distance of the group average to the target and the dispersion of individuals from the group average. (4) Using these indices in the two-step optimization process to shift the mean performance of the population to target and reduce variability in performance. (5) Apply information obtained from experiment to generate multivariate process capability estimates.; In the case study, significant gains in fuel economy and vehicle performance were demonstrated. These gains were brought about, without introducing any new manufacturing or product technology, by adapting this method to identify and optimize design nominal.; The application of the T2-statistic allows the use of univariate tools in multiple objective problems. Furthermore, the decomposition of T2O into T2M and T2D substitutes a complex multivariate optimization process with the simpler two-step procedure. Finally, using information from the experiment, a multivariate process capability estimates for the design can be made prior to hardware fabrication.