Robust Parameter Design: Modeling, Analysis and Layout Techniques

9704649吴建富J.稳健参数设计采用统计方法和工程思想来提高产品或过程的质量，使其对不可控的变化不那么敏感。正如许多工业案例研究所证明的那样，它对制造业产生了巨大影响。主要重点是研究参数设计实验数据的建模和分析，以及有效实验规划的布局技术。首先，考虑了具有静态特性的鲁棒设计的三个方面：稳健因子设置的理论特征及其与常用图形方法的关系2. 使用非常一般的损失函数，特别是在无法确定调整因子的情况下，由于质量差造成的损失；和3。评估收集有序分类数据的测量系统。与最后一个方面相关的一个问题是，当一些数据响应被错误分类时，如何对来自设计实验的有序分类数据进行建模和分析。接下来，研究了具有功能特征（即响应是函数关系而不是单个值）的鲁棒参数设计的两个方面：基于广义信噪比的方法与采用两阶段建模的响应函数建模方法的比较，以及实验中退化曲线的建模和分析，以提高鲁棒可靠性。最后，研究了静态特性参数设计和信号响应系统参数设计的有效规划。新的理论和计算工具被用来解决这个问题。鲁棒参数设计采用统计方法和工程思想来提高产品或过程的质量，使其对不可控的变化不那么敏感。许多工业案例证明了该方法的成功实施。研究人员研究了鲁棒参数设计在更复杂情况下的应用，从而扩展了鲁棒参数设计的范围。稳健参数设计使用实验，其两个主要问题是如何正确选择实验条件（即实验计划）和如何建模和分析数据（即分析），以识别重要因素并推荐这些因素的最佳设置。在第一部分中，(I)获得了与图形方法相关的稳健因子设置的理论特征，（ii）开发了新的策略，以最大限度地减少由非常一般的损失函数测量的质量差造成的损失，（iii）评估和分析收集有序分类数据的测量系统。在第二部分中，(i)比较了分析信号响应系统的两种方法（例如，气体表，其中信号是油箱中的气体量，响应是气体表的读数）；（II）研究了分析退化曲线的一般模型和方法（例如，荧光灯强度随时间降低）。使用退化数据是有策略的，因为有关可靠性的信息可以更快、更有效地获得。第二部分的研究成果将在一个汽车项目上进行检验。第三部分研究了参数设计问题中实验的高效经济规划问题。

9704649 Chien-Fu J. Wu Robust parameter design employs statistical methods and engineering ideas to improve the quality of a product or process by making it less sensitive to uncontrollable variation. It has had a big impact on manufacturing as demonstrated by many industrial case studies. The main focus is to study modeling and analysis of data from parameter design experiments and layout techniques for efficient experimental planning. First, three aspects of robust design with static characteristics are considered: 1. theoretical characterizations of robust factor settings and their relation to commonly used graphical methods; 2. loss due to poor quality using very general loss functions, especially where no adjustment factor can be identified; and 3. assessment of measurement systems that collect ordered categorical data. A related problem to the last aspect is the modeling and analysis of ordered categorical data from designed experiments when some of the data responses are misclassified. Next, two aspects of robust parameter design with functional characteristics (i.e., whose response is a functional relation, instead of a single value) are studied: comparison of generalized signal-to-noise ratio based method to a response function modeling method that employs a two-stage modeling, and the modeling and analysis of degradationcurves from experiments for robust reliability improvement. Finally, efficient planning of experiments is studied for parameter designswith static characteristics and with signal-response systems. Novel theoretical and computational tools are employed to solve this problem. Robust parameter design employs statistical methods and engineering ideas to improve the quality of a product or process by making it less sensitive to uncontrollable variation. It has been successfully implemented as demonstrated by many industrial case studies. The investigators study the use of robust parameter design in more complicated situations than considered to date, thereb y, extending the scope of robust parameter design. Robust parameter design uses experiments whose two major issues are how to properly choose the experimental conditions (i.e., experimental plans) and how to model and analyze the data (i.e., analysis) to identify the important factors and recommend the optimal settings for these factors. In Part I, (i) theoretical characterizations of robust factor settings are obtained and related to graphical methods, (ii) new strategies are developed for minimizing loss due to poor quality as measured by very general loss functions, (iii)assessment and analysis of measurement systems that collect ordered categorical data. In Part II, (i) two methods of analyzing signal-response systems (e.g., gas gauge where the signal is the amount of gas in the gas tank and the response is the reading from the gas gauge) are compared (ii) general models and methods for analyzing degradation curves are investigated, (e.g., fluorescent light intensity decreases over time). Using degradation data is strategic because information about reliability can be obtained sooner and more efficiently. The research results in Part II will be tested on an automobile project. In Part III, the problem of efficient and economic planning of experiments is studied for parameter design problems.