基于多保真度模型的产品设计优化决策方法研究

In real industrial applications, there exist many product design optimization problems that involve highly complex systems that are difficult to be modeled and studied analytically. Simulation is usually applied to simulate the designed system and evaluate different designs. The high-fidelity simulation model can represent the system accurately but the running time is long. In addition, complex product design usually contains many design dimensions. So the number of alternative designs is large. Because of these challenges, solving the product design optimization problem for a large-scale complex system is a very time-consuming process. Therefore, besides the accuracy, the computational efficiency is also an utterly important issue for performing optimization on complex systems design problem. .Different simulation models or analytical models can be built for the same systems. The low fidelity models only provide rough approximations to the systems’ performance but they can be run very easily and quickly. By utilizing the advantages of low-fidelity model’s time cost and high-fidelity model’s accuracy, this project aims to propose an efficient and high quality framework for complex system design optimization under multi-fidelity models. In this project, we intend to seek for a foundation of models evaluation, optimization search, and computing resource allocation. The key contributions of this project are: i) utilizing low- and high-fidelity models for finding the best design of the system more efficiently; ii) a novel ordinal transformation method will be developed to study the relationship between low- and high-fidelity models; iii) maximizing the efficiency of identifying best design alternative under limited computing time resources.

随着当今产品设计所涉及的系统日渐复杂，产品设计优化问题目前常使用仿真方法来对所设计的系统进行模拟进而评估不同设计方案的优劣。高保真度的仿真模型可以对系统进行高精度的模拟，但是运行需要很长时间。此外，复杂产品会有多个设计维度，其可行的设计方案数目将非常庞大。因此，计算效率是产品设计优化中非常重要的一个问题。.同一个系统的低保真度模型对系统的表现评估会较粗糙，但是所需的计算时间成本会很低。此项目合理利用低保真度模型运行时间短和高保真模型高精确度的优势，提出提高产品设计优化决策效率的方法。该项目主要从高低保真度模型的关系、最优设计方案的搜索算法以及不同设计方案之间的时间分配这三方面来进行研究。主要贡献包括三方面：（1）利用低保真度模型来极大提高选择出最优的产品设计方案的效率；（2）创新性地提出了次序转换方法来分析高低保真度模型之间的关系；（3）在有限的时间资源约束下最大化正确选择最优方案的概率。

此项目主要研究多保真度模型下的优化设计问题。高保真度的模型可以对系统进行高精度的模拟，但是运行需要很长时间。低保真度模型对系统模拟评估精度差有误差，但是计算量较小。本项目通过合理利用低保真度模型运行时间短和高保真模型高精确度的优势，提出提高产品设计优化决策效率的方法。. 该项目主要从高低保真度模型的关系、最优设计方案的搜索算法以及不同设计方案之间的时间分配这三方面来进行研究。主要研究结果包括四方面：（1）基于降维思想提供了建立合理低保真度模型的方法；（2）提出适用于多保真度模型下的优化算法，即改进粒子群算法；（3）开发了在有限的时间资源约束下改进求解效率的PSO-OCBA算法；（4）将该项目的理论结果以实际案例为背景进行应用检验，算法表现良好。此项目的研究为解决大规模优化设计中计算量巨大的问题提供了高效的理论解决方法，并且在实际应用领域有着较好的应用前景。