A Bayesian Treatment of Uncertainty in Simulation-Based Methods for Enhancing Process and Product Robustness

摘要提高过程和产品稳健性的基于模拟的方法中不确定性的贝叶斯处理这笔赠款提供资金，用于创建一种贝叶斯方法，用于在使用基于模拟的方法增强制造过程和产品的稳健性时管理各种形式的不确定性。将处理的三种不确定性是参数不确定性、模拟不确定性和模型不确定性。参数不确定性的一个例子是冲压过程中材料特性的变化。模拟不确定性源于对输入变量组合的数量的限制，在这些输入变量组合处，人们可以进行计算昂贵的模拟运行。模型的不确定性是由于模拟输出与实际世界之间的差异造成的。将使用贝叶斯框架来定量表示所有三种形式的不确定性的影响，即它们对稳健设计目标的影响。这种面向目标的表示将构成支持关键决策的贝叶斯方法的基础，例如指导模拟和物理实验为优化设计提供最大信息，确定当前信息是否足以终止模拟并自信地优化设计，以及确保设计解决方案真正产生对所有三种形式的不确定性的健壮性。基于物理实验的可靠设计是坚定的实践。然而，为了缩短开发周期或当物理实验不切实际时，计算机模拟日益成为物理实验的重要替代或补充。如果成功，这项研究的结果将为有效实现基于计算机模拟的稳健设计优化提供急需的工具。由于这种研究不局限于特定类型的仿真代码，并且允许用户在有或没有约束的情况下从各种概率目标函数中进行选择，因此有望得到广泛的应用。为使分析结果可视化而开发易于解释的图形显示将有助于该方法的实施，扩大其预期影响。

ABSTRACTA Bayesian Treatment of Uncertainty in Simulation-Based Methods for Enhancing Process and Product Robustness This grant provides funding to create a Bayesian methodology for managing various forms of uncertainty when using simulation-based methods for enhancing the robustness of manufacturing processes and manufactured products. Three types of uncertainty that will be treated are parameter uncertainty, simulation uncertainty, and model uncertainty. An example of parameter uncertainty is variation in material properties in stamping processes. Simulation uncertainty results from limitations on the number of input variable combinations at which one may conduct computationally expensive simulation runs. Model uncertainty results from differences between the simulation output and the physical world. A Bayesian framework will be used to quantitatively represent the effects of all three forms of uncertainty, in terms of their impact on the robust design objective. This objective-oriented representation will form the basis for a Bayesian methodology for supporting critical decision making, such as guiding the simulation and physical experiments to provide the greatest information for optimizing the design, deciding whether current information is sufficient to terminate simulation and confidently optimize the design, and ensuring that the design solution truly results in robustness to all three forms of uncertainty.Robust design based on physical experimentation is firmly established practice. However, to reduce development cycle time or when physical experiments are impractical, computer simulations are increasingly important replacements for or supplements to physical experimentation. If successful, the results of this research will provide much needed tools for efficiently achieving robust design optimization based on computer simulation. Because this research is not restricted to a particular type of simulation code and allows the user to choose from a variety of probabilistic objective functions, with or without constraints, it is expected to find widespread application. Development of easy-to-interpret graphical displays for visualizing the analytical results will facilitate implementation of the methodology, broadening its expected impact.