Collaborative Research: Model-Based Multidisciplinary Dynamic Decisions in Design

航天飞行器、汽车和先进材料系统等工程系统的复杂性已经达到了一个临界点，对现有的设计和分析方法提出了挑战。应用于传统范例中的简化设计模型不足以捕捉复杂的系统行为。此外，高保真的计算机仿真模型和实验测试由于成本较高，在许多情况下不能完全应用。因此，非常需要系统地融合来自多个来源的信息，包括具有多个保真度的仿真模型，并决定如何最好地在设计过程的每个阶段进行进一步的仿真。这项研究将创建一个新的决策框架来满足这一需求，并指导复杂工程系统的设计。由此产生的方法预计将通过减少开发周期后期发生的未发现问题的发生，并减少设计过程所需的预算和进度，从而增加航空航天、汽车、能源和消费电子等行业的设计过程的价值。该项目还将提供跨越航空航天、机械和工业工程的跨学科研究、培训和教育。这项研究的智力意义在于将复杂系统的设计视为一个可以建模为随机离散时间动态系统的信息寻求和知识生成(学习)过程。将信息论和决策科学结合起来进行设计决策，不仅涉及系统属性的选择，而且涉及设计过程中后续信息寻求行为的选择。将建立一个总体贝叶斯空间随机过程建模框架，以融合来自多保真度模拟和实验的不同信息和不确定性量化，其中模型的保真度可以明确排序(分级)或不分级(非分级)。将建立以多学科统计敏感性分析和多学科不确定性分析为基础的方法，以管理在不同保真度和学科之间融合信息所固有的耦合和复杂性，同时在分布式分析中保持学科自主权。动态决策框架将为许多不同类型的不确定性提供统一的解释，并且基于期望效用理论的决策函数被用来指导后续的设计行动。这项研究的另一个关键贡献是开发了基于启发式的策略，用于管理解决令人望而生畏的最优决策问题的复杂性。该框架将在涉及分布式电力推进飞机概念设计的试验台问题上进行评估。

The complexity of engineered systems such as aerospace vehicles, automobiles, and advanced materials systems has reached a tipping point that challenges existing design and analysis methods. Simplified design models, applied in traditional paradigms, are inadequate for capturing complex system behaviors. Moreover, high-fidelity computer simulation models and experimental tests cannot be fully applied in many situations, due to their high costs. As a result, there is a great need for systematically fusing information from multiple sources, including simulation models with multiple levels of fidelity, and for deciding how best to conduct further simulations at each stage of the design process. This research will create a new decision-making framework to address this need and to guide the design of complex engineered systems. The resulting method is expected to increase the value of the design process in industries such as aerospace, automotive, energy, and consumer electronics by reducing the occurrence of undiscovered problems that occur late in a development cycle and decreasing the budget and schedule required for the design process. The project will also provide interdisciplinary research training and education across aerospace, mechanical, and industrial engineering.The intellectual significance of this research is to approach the design of a complex system as an information-seeking and knowledge-generation (learning) process that can be modeled as a stochastic discrete-time dynamical system. Information theory and decision science are integrated to make design decisions that involve not only selection of system attributes, but also choices about subsequent information-seeking actions in a design process. An overarching Bayesian spatial random process modeling framework will be established to fuse heterogeneous information from multifidelity simulations and experiments with uncertainty quantification, where the fidelity of models can be either clearly ranked (hierarchical) or not (nonhierarchical). Approaches based on multidisciplinary statistical sensitivity analysis and multidisciplinary uncertainty analysis will be established for managing the couplings and complexity inherent in fusing information across fidelities and disciplines, while maintaining disciplinary autonomy in distributed analyses. The dynamic decision making framework will provide a uniform accounting for many different types of uncertainties, and decision functions grounded in expected utility theory are formulated to guide subsequent design actions. A further critical contribution of this research is to develop heuristic-based strategies for managing the complexity in solving the daunting optimal decision making problem. The framework will be assessed on a testbed problem involving the design of a distributed electric propulsion aircraft concept.