基于近似模型的两级集成系统协同优化方法研究

发布时间：2018-06-07 11:01

本文选题：多学科设计优化 + 优化过程　；参考：《华中科技大学》2012年硕士论文

【摘要】：复杂产品的设计通常涉及到众多学科，而且学科之间存在复杂的耦合关系，能够解决复杂、强耦合设计问题的多学科设计优化方法（MDO）的作用越来越突出。MDO优化过程是多学科设计优化中最核心、最重要的内容，直接决定了MDO技术在具体的工程优化问题中的应用可靠性。国内外学者已经对MDO优化过程进行了较深入的研究，并取得了一定的成果。但是由于产品的多样性和复杂性，还没有形成一种能广泛应用于各种MDO问题的方法。同时，近似模型技术因其既能保证一定的精度，又能大幅度降低重复调用精确仿真模型所需的计算成本，已成为研究复杂系统MDO问题时必不可少的工具。本文主要对MDO优化过程和近似模型进行研究，寻求一种能更快更准找到全局最优方案的高效、高可靠性的MDO优化方法。首先，本文对MDO优化过程和近似模型技术的国内外研究现状进行了调研，从分析系统优化与子系统优化之间的关系出发，探讨了一般分解-协调优化过程的迭代流程，并对几种主要的多级MDO优化过程进行了研究，从分解技术、收敛性、求解效率等多方面分析比较了它们之间的优劣。其次，本文在已有的研究基础之上，对其中综合优化性能较好的两级集成系统协同优化(BLISCO)方法进行了深入研究，针对该方法所存在的每次迭代需要调用高计算量的精确仿真分析、学科优化易受数值噪声影响等问题，对其进行了改进，提出了基于近似模型技术的BLISCO-AM方法，并给出了该方法的算法结构和优化流程。本文通过一个强耦合的非线性优化问题对该改进方法的可行性进行了验证。与原来的BLSCO方法相比，该改进方法大大减少了子系统分析次数，提高了求解效率。同时，本文还对改进方法中的一致性约束进行了研究，通过对比采用不同允许容差ε获得的优化结果，，证实了在算法中用不等式约束代替严格等式约束的可行性和有效性。最后，在以上研究成果的基础上，针对齿轮减速箱优化问题，对改进的BLISCO-AM方法中采用不同近似模型技术的不同优化效果进行了研究。试验结果表明采用响应面模型的BLSCO-AM方法和采用Kriging模型的BLISCO-AM方法均获得了较好的收敛效果，且大大降低了计算量，进一步验证了该改进方法的可行性和高效性，并且构造的近似模型精度越高，优化效果越好。研究还表明并非对于所有的问题，Kriging模型的近似精度都高于响应面模型，因此应根据具体的问题选择适合的近似技术。
[Abstract]:The design of complex products usually involves many disciplines, and there are complex coupling relationships between them, which can solve the problem of complexity. The role of multi-disciplinary design optimization method in strongly coupled design problem is more and more prominent. MDO optimization process is the core and most important content of multidisciplinary design optimization, which directly determines the reliability of the application of MDO technology in specific engineering optimization problems. Scholars at home and abroad have done more in-depth research on MDO optimization process, and have achieved certain results. However, due to the diversity and complexity of products, there has not been a method that can be widely used in various MDO problems. At the same time, the approximate model technology has become an indispensable tool to study the MDO problem of complex systems because it can not only guarantee a certain accuracy but also reduce the computational cost of repeating the accurate simulation model. In this paper, the optimization process and approximate model of MDO are studied to find an efficient and reliable MDO optimization method which can find the global optimal scheme more quickly and accurately. Firstly, this paper investigates the research status of MDO optimization process and approximate model technology at home and abroad. Based on the analysis of the relationship between system optimization and subsystem optimization, the iterative process of general decomposition-coordination optimization process is discussed. Several main multistage MDO optimization processes are studied and their advantages and disadvantages are analyzed and compared from decomposition technology convergence efficiency and so on. Secondly, on the basis of the existing research, this paper makes a deep research on the cooperative optimization method of two-level integrated system, which has better performance of integrated optimization. In order to solve the problems existing in this method, such as accurate simulation analysis with high computational load and subject optimization easy to be affected by numerical noise, the BLISCO-AM method based on approximate model technology is proposed. The algorithm structure and optimization flow of the method are also given. The feasibility of the improved method is verified by a strongly coupled nonlinear optimization problem. Compared with the original BLSCO method, the improved method greatly reduces the times of subsystem analysis and improves the efficiency of solution. At the same time, the consistency constraints in the improved method are studied in this paper. By comparing the optimization results obtained by using different allowable tolerances 蔚, the feasibility and effectiveness of replacing strict equality constraints with inequality constraints in the algorithm are proved. Finally, on the basis of the above research results, aiming at the problem of gear reducer optimization, the different optimization effects using different approximate model techniques in the improved BLISCO-AM method are studied. The experimental results show that both the BLSCO-AM method based on response surface model and the BLISCO-AM method using Kriging model have better convergence effect, and greatly reduce the computational complexity. The feasibility and efficiency of the improved method are further verified. And the higher the precision of the approximate model is, the better the optimization effect is. The results also show that not all the Kriging models have higher accuracy than the response surface model, so we should choose the appropriate approximation technology according to the specific problems.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH122

【参考文献】