Optimizing product design by application of experimental design for computer simulation

计算机模拟在最近的研究和开发阶段，产品设计阶段等方面做出了很大的贡献，在降低成本，早期开发方面。计算机模拟实验之所以受欢迎，是因为它不需要进行物理实验。在本研究中，我们提出了一种优化产品设计的一般方法，其中心是实验设计，这是一系列用于数据收集和分析的统计技术。本研究的主要课题之一是如何验证正在开发的计算机仿真模型。通过与在汽车行业工作的从业人员的讨论，获得了这个主题的重要性。通过讨论，得出了在汽车行业的验证实践，如成本约束、时间进度等。根据实际需要，提出了统计必要性。在考虑统计必要性的前提下，给出了一种序列实验算法和实验设计准则。此外，我们还重点研究了从大量候选因素中找出重要因素的筛选技术和描述响应与因素之间关系的近似技术，以便利用已开发的计算机模拟器。在筛选方面，我们建议在传统的分数因子设计和统计检验方法中，当因子过多时，应用过饱和设计和逐步选择。传统方法在实验中可容纳的因素数量上存在限制，而本文提出的方法可以包含比传统方法更多的因素。在近似方面，均匀设计适合于描述响应与因素之间的复杂关系，而复合设计和二阶模型是常用的近似方法。与传统方法相比，该方法可以灵活地检验各种模型来近似响应与因素之间的关系。少

Computer simulation has contributed the recent research and development stage, product design stage and so forth very much in terms of cost reduction, early development. The reason of the popularity is that experiments by computer simulation do not require performing the physical experiments. In this research, we has proposed a general approach for optimization of the product design whose central is design of experiments that is a series of statistical techniques for data collection and analysis. One of the main topics in this research is a technique to validate the developing computer simulation model. The importance of this topic was obtained by discussion with practitioners who work in automotive industry. Through the discussion, the validation practices in automotive industry are obtained, such as cost constraints, time schedule and so forth. Based on the practical requirements, statistical necessity has been formulated. Furthermore, an algorithm for sequential experiments and the … More criteria for design of experiments have been obtained with considering the statistical necessity. In addition, we focus on screening techniques to find important factors from a lot of candidate factors and approximation techniques to describe the relation between responses and factors in order to utilize the developed computer simulator. As regard the screening, we propose an application of supersaturated design and stepwise selection when there are too many factors for the traditional approach by fractional factorial design and statistical tests. While the traditional approach has the limitation of number of factors that can be accommodated in experiments, proposed approach can include more factors than the traditional approach. As regard the approximation, it was found that uniform design is suitable to describe a complicated relation between responses and factors while composite design and second order model is a popular approach. The proposed approach has a flexibility to examine various models to approximate the relation between responses and factors comparing with the traditional approach. Less