Simulation Uncertainty in Multidisciplinary Design

该补助金为开发基于决策的设计框架提供资金，用于设计复杂的工程系统，如汽车，飞机或消费品。 这些工程系统的设计涉及计算机仿真设计工具的广泛使用。 每个计算机模拟模型都是现实的抽象，并且具有与其性能预测相关的一些不确定性。 本研究将探讨这些不确定性如何通过一个多学科的系统，包括许多学科特定的仿真工具传播。 本研究的一个目标是能够量化这些传播的不确定性，用于基于决策的设计框架。 这项调查将评估技术可从实验数据分析，估计在复杂的工程系统中的最坏情况下传播的不确定性，并调查的效用进行Monte Carlo模拟的响应面近似的系统性能量化的传播的不确定性的分布。 在量化了复杂系统中传播的不确定性之后，这项研究工作的第二阶段将致力于实施基于决策的设计框架。 基于信赖域管理的增广拉格朗日方法的优化算法，用于并行子空间设计/优化，将用于驱动基于决策的设计框架。 这种优化将基于系统性能的人工神经网络响应面近似。 在随机性能约束条件下，优化器将最大化产品总生命周期成本的净收入的预期效用。 预计这项研究的结果将导致产品性能的改善，以及在降低成本和风险的情况下减少产品开发时间。 一个基于决策的设计框架的发展将表明，设计师可以有效地管理与基于仿真的新产品设计的不确定性和风险。

This grant provides funding for the development of a decision based design framework for the design of complex engineering systems such as automobiles, aircraft or consumer products. The design of these engineering systems involves extensive use of computer simulation design tools. Each computer simulation model is an abstraction of reality and has some uncertainty associated with its performance predictions. This research will investigate how these uncertainties propagate through a multidisciplinary system comprised of many discipline-specific simulation tools. A goal of this research is to be able to quantify these propagated uncertainties for use in a decision-based design framework. This investigation will evaluate techniques available from experimental data analysis for estimating the worst case propagated uncertainty in complex engineering systems and investigate the utility of performing Monte Carlo simulations on response surface approximations of system performance for quantifying the distributions of the propagated uncertainties. Having quantified propagated uncertainties in complex systems, the second phase of this research effort will be devoted to the implementation of a decision-based design framework. An optimization algorithm based on a trust region managed augmented Lagrangian approach for concurrent subspace design/optimization will be used to drive the decision-based design framework. This optimization will be based on artificial neural network response surface approximations of the system performance. The optimizer will maximize the expected utility of the net revenue for the total life cycle cost of the product subject to stochastic performance constraints. It is anticipated that the results of this research will lead to improved product performances as well as reduced product development times at reduced cost and risk. The development of a decision-based design framework will demonstrate that designers can effectively manage both the uncertainty and risk associated with the simulation based design of new products.