Data driven computational frameworks for robust design optimization of complex engineering systems

基于模拟的预测工具彻底改变了工程设计实践，使设计人员能够在建造和测试物理原型之前提高系统性能和安全性。然而，为了将高保真仿真工具应用于设计复杂的工程系统，仍有许多计算挑战需要解决。拟议的研究计划将提供有效的计算方法和软件框架，使工程师能够在有限的计算预算中优化复杂工程系统的性能和健壮性。这项工作的重点将是开发新的贪婪函数逼近策略，以构建计算高效的高保真仿真模型的代理模型。这项工作还将包括线性和非线性降维策略的制定，这些策略将使贪婪代理建模算法能够应用于工程实践中通常遇到的高维、大规模模拟数据库。将利用为代理建模开发的算法，开发用于稳健设计的数据驱动的优化框架。在本研究项目中开发的计算方法将在航空发动机设计和气动外形优化问题上进行测试和验证。这项研究项目的主要成果将是用于构建代理模型的新的贪婪算法、用于稳健设计的代理辅助优化策略，以及将使复杂的真实世界工程系统的设计成本显著降低的通用软件工具包。

Simulation-based predictive tools have revolutionized engineering design practice, allowing designers to improve system performance and safety before physical prototypes are built and tested. However, a number of computational challenges remain to be addressed in order to apply high-fidelity simulation tools to design complex engineering systems. The proposed research program will deliver efficient computational methods and a software framework that will enable engineers to optimize the performance and robustness of complex engineering systems on a limited computational budget. The focus of this work will be on the development of novel greedy function approximation strategies for constructing computationally efficient surrogate models of high-fidelity simulation models. This work will also include the formulation of linear and nonlinear dimensionality reduction strategies that will enable the application of greedy surrogate modeling algorithms to high-dimensional, large-scale simulation databases typically encountered in engineering practice. Data-driven optimization frameworks for robust design will be developed leveraging the algorithms developed for surrogate modeling. The computational methods developed during this research project will be tested and validated on aeroengine design and aerodynamic shape optimization problems. The key deliverables from this research project will be novel greedy algorithms for constructing surrogate models, surrogate-assisted optimization strategies for robust design, and a general-purpose software toolkit that will enable significant reductions in the design cost for complex real-world engineering systems.