【作者】 李伟；

【导师】 高亮； 肖蜜； Akhil Garg； 李培根；

【作者基本信息】 华中科技大学 ， 机械工程， 2020， 博士

【摘要】 目前参数不确定性下多学科设计优化(MDO)已经取得了一系列的研究成果,但现有的研究往往未考虑模型的不确定性。大量的事实表明复杂机械系统的计算机仿真模型和元模型(也称为代理模型)都存在不确定性。由于复杂机械系统具有多参数、多约束和强耦合等特点,不确定性下的MDO研究必须考虑参数和模型之间的相互影响。目前国内外关于综合考虑参数和模型不确定性的MDO研究非常少见。因此,本文开展了考虑参数和模型不确定性的多学科稳健设计优化(MRDO)方法研究。本文的主要工作归纳如下:(1)提出了参数不确定的MRDO方法。该方法通过最大变差分析法(MVA)进行参数不确定分析,建立了系统和子系统的内外嵌套优化框架,利用该框架分别求解系统和子系统的稳健最优解。利用目标级联法(ATC)对复杂系统进行划分,实现了系统级和子系统之间的协调,从而保证了系统和子系统的稳健解的一致性。通过MDO数学算例和心脏偶极优化实例,对该方法进行了验证。该方法为研究参数不确定的MRDO问题提供了一种新的尝试。MVA与ATC方法相结合的优化框架简单易行,为复杂系统高效优化提供了便利。(2)提出了参数和模型不确定的MRDO方法。该方法利用区间方法量化参数不确定性。借助贝叶斯方法,量化模型不确定性,即在计算机模型中加入偏差函数,以最大程度地消除计算机模型与实际物理系统输出的偏差。通过一种高效的协同模型采样技术,获得足够多的多学科可行样本,分别建立了计算机模型和偏差函数的高斯过程(GP)模型。构建了考虑参数和模型不确定性的MRDO框架。通过MDO数学算例和电源转换器设计实例,对该方法进行了验证。该方法考虑了多学科系统中的模型不确定性,并通过协同模型,避免了复杂的多学科计算,提高了计算效率。(3)提出了参数和元模型不确定的MRDO方法。探讨了参数和元模型不确定性的复合效应对系统性能的影响。利用协同模型,获得满足多学科可行的样本集。利用这些样本,构建了计算机模型的GP元模型,并对其进行评估以保证需要的精度。采用蒙特卡洛模拟(MCS)方法,对参数和元模型不确定性的复合效应进行量化。建立了考虑参数和元模型不确定性的MRDO优化框架。通过MDO数学算例和减速器设计实例,对该方法进行了验证。该方法探讨了多学科系统中元模型的不确定性,提高了优化结果的稳健性。(4)提出了参数、模型和元模型不确定的MRDO方法。该方法利用区间方法量化参数不确定性。借助贝叶斯方法,量化模型不确定性。通过MCS方法,量化MRDO中的参数、模型和元模型不确定性的综合影响。搭建了考虑参数、模型和元模型不确定下的MRDO平台。通过MDO数学算例和薄壁压力容器设计实例,对该方法进行了验证。该方法综合考虑了复杂系统中参数、模型和元模型不确定性,为工程实际中复杂系统稳健优化提供了有益的探索和尝试。(5)以电动汽车液冷电池热管理系统(BTMS)的设计优化问题为背景,采用本文所提方法,分别对方形BTMS和圆柱形BTMS进行了MRDO研究。探讨了BTMS优化中的参数、模型和元模型不确定性。本文所提MRDO方法在工程实例中的成功应用,可为BTMS的设计优化问题提供一些借鉴和参考。更多还原

【Abstract】 At present,multidisciplinary design optimization(MDO)has achieved a series of research results under parameter uncertainty.However,existing research often fails to consider the model uncertainty.A large number of facts indicate that there are uncertainties in the computer simulation model and metamodel(also called surrogate model)of complex mechanical systems.Due to the multi-parameter,multi-constraint,and strong coupling characteristics of complex mechanical systems,the uncertain MDO research must consider the interaction between parameter and model uncertainties.At present,domestic and foreign MDO studies on comprehensive consideration of parameter and model uncertainties are very rare.Therefore,this paper carried out a multidisciplinary robust design optimization(MRDO)method considering the parameter and model uncertainties.The main work of this article is summarized as follows:(1)A MRDO method considering parameter uncertainty is proposed.This method uses the maximum variation analysis(MVA)method to perform parameter uncertainty analysis,establishes an internal and external nested optimization framework for the system and subsystems,and uses the framework to solve the robust and optimal solutions of the system and subsystems,respectively.The analytical target cascading(ATC)is used to divide and coordinate complex systems to achieve coordination between the system and the subsystems,thereby ensuring the consistency of the robust solutions of the system and subsystems.The proposed method is verified by an MDO mathematical example and a heart dipole optimization problem.This method provides a new attempt to study MRDO problems with parameter uncertainty.The optimization framework combining MVA and ATC methods is simple and easy,which facilitates the efficient optimization of complex systems.(2)A MRDO method with parameter and model uncertainties is proposed.This method uses the interval method to quantify parameter uncertainty.Bayesian method is used to quantify the model uncertainty,i.e.,adding a bias function to the computer model to eliminate the deviation of the computer model from the actual physical system output to the greatest extent.A sufficient number of multidisciplinary feasible samples are obtained through an efficient collaboration model.Then Gaussian process(GP)models of the computer model and the bias function are established using the obtained samples,respectively.A MRDO framework considering parameter and model uncertainties is constructed.The method is verified by an MDO mathematical example and a power converter design problem.This method considers the model uncertainty in multidisciplinary systems,and avoids complex multidisciplinary calculations through the collaboration model,improving computational efficiency.(3)A MRDO method with parameter and metamodeling uncertainties is proposed.The effects of the combined effects of parameter and metamodel uncertainties on system performance are discussed.The collaboration model is used to obtain samples that meet the requirements of multidisciplinary characteristics.GP metamodels of the computational model are constructed by obtained samples.Then GP metamodels are evaluated to ensure the required accuracy.Monte Carlo simulation(MCS)method is used to quantify the compound effect of parameter and the metamodeling uncertainties.A MRDO optimization framework considering parameter and metamodeling uncertainties is established.An MDO mathematical example and reducer design problem verify the method.This method explores the metamodeling uncertainty in a multidisciplinary system and improves the robustness of the optimization results.(4)A MRDO method with parameter,model,and metamodeling uncertainties is proposed.This method uses interval method to quantify parameter uncertainty.Bayesian method is taken to quantify model uncertainty.The MCS method is used to quantify the combined effects of parameter,model,and metamodeling uncertainties in MRDO.An MRDO platform is established under the parameter,model,and metamodeling uncertainties.The method is verified by an MDO mathematical example and thin-walled pressure vessel design example.This method comprehensively considers the parameters,model,and metamodel uncertainties in complex systems,and provides useful explorations and attempts for robust optimization of complex systems in engineering practice.(5)Taking the design optimization problem of electric vehicle liquid-cooled battery thermal management system(BTMS)as the background,the MRDO study of square BTMS and cylindrical BTMS is carried out by using the proposed method in this paper.The parameter,model,and metamodel uncertainties in BTMS optimization are discussed.The successful application of the proposed MRDO approach in the engineering example provides some reference for the design optimization of BTMS.更多还原