Optimization of complex mechanics simulations with object-oriented software designs
Optimization of complex mechanics simulations with object-oriented software designs
复制标题
使用面向对象的软件设计优化复杂力学模拟
DOI：
10.2514/6.1995-1433
复制
发表时间：
1995
影响因子：
0.5
通讯作者：
Sandia National Laboratories
中科院分区：
文献类型：
--
作者：
M. S. Eldred;D. Outka;W. Bohnhoff;W. R. Witkowski;V. J. Romero;E. R. Ponslet;K. S. Chen;Sandia National Laboratories
将优化应用于计算模型的好处是众所周知的，但迄今为止其应用范围仍然有限。这项工作试图通过开发和应用能够处理与复杂模拟代码相关的计算困难的高级优化策略来扩展优化算法可以轻松应用的学科领域。为了实现这一目标，正在持续开发灵活的软件框架，以便将优化技术应用于广泛的工程应用，包括那些计算费用高的应用和设计空间不平滑的应用。使用 C++ 进行面向对象的软件设计已被用作提供灵活、可扩展且强大的多学科工具包的工具，该工具包建立了围绕计算密集型模拟进行参数优化的协议。面向对象的方法非常适合处理这种大型软件任务，其中无数种类的优化算法、近似技术和混合策略必须与广泛的分析功能通用地接口。使用该软件进行优化的演示在流体力学、传热学、非线性固体力学及其组合中进行。给出了最佳结果以及优化过程中学到的技术经验教训。 *结构动力学部技术人员 (SMTS) 高级会员、AIAA 会员。 fSMTS，导航制导部，AIAA 成员。 SMTS，结构动力学部，AIAA 成员。 §SMTS，智能系统部 W.0。信箱 5800，阿尔伯克基，新墨西哥州 87185-0439，美国。这项工作在桑迪亚国家实验室进行，并得到美国能源部的支持，合同号为 DE-AC04-94AL85000。本文被宣布为美国政府的作品，在美国不受版权保护。简介 在流体力学、结构动力学、传热、非线性大变形力学、制造和材料加工以及许多其他工程领域中开发的计算方法可以极大地帮助理解它们所模拟的复杂物理系统。通常，希望利用这些模拟作为虚拟原型来改进或优化特定系统的设计。本文描述的工作目标是通过为这类计算方法提供通用优化能力来增强它们的实用性。这填补了计算模拟中的一个关键空白。一个重大影响将是提高计算方法作为模拟制造过程、工具或产品的设计工具的实用性。优化目标可用于最小化重量或缺陷，或最大化性能、可靠性、吞吐量、可重构性、敏捷性或设计稳健性（对非标称参数值不敏感）。确定这些最佳解决方案的系统、快速方法将带来更好的设计、降低硬件费用、缩短设计周期并缩短上市时间。尽管当前的计算专业知识和能力很广，但计算方法的影响被削弱，并最终由于无法使用这些方法找到“最佳”解决方案而受到限制。为了实现这些目标，我们针对广泛的计算方法的需求，以提供通用的优化能力。以前的优化工作的学科范围是有限的，并且许多工作都集中在行为良好（即光滑、凸）问题上[1]，[2]。用于这些表现良好的问题的标准优化技术在应用于具有高计算时间和复杂性的问题时表现不佳，包括那些具有严重非线性、不连续和多模态参数空间的问题。解决这些问题扩大了可以实现优化优势的应用范围。技术问题。优化技术在一组约束条件下迭代地操纵一组参数以实现目标函数的极值。优化与复杂计算方法的耦合很困难，并且优化算法通常无法有效收敛（如果有的话）。困难来自于许多计算方法所共有的以下特征： 1.用一组参数完成单个函数评估所需的时间很长。因此，最大限度地减少功能评估的数量至关重要。 2. 目标函数和约束函数的解析导数（相对于参数）经常不可用。因此，基于灵敏度的优化方法依赖于数值生成的梯度，这需要对每个标量参数进行额外的函数评估。 3.参数可以是连续的或离散的，或两者的组合。 4. 目标函数和约束函数可能不平滑或表现不佳；即，响应曲面在参数空间的某些区域中可以是非线性的、不连续的，甚至是未定义的。多个局部极值（多模态）的存在是常见的。 5. 嵌入式迭代方案中的收敛容差会在函数评估响应面中引入不确定性（噪声），从而导致数值梯度不准确。 6. 每个功能评估可能需要“初步猜测”。函数评估对初始猜测的依赖可能会导致响应面出现额外的不确定性。此外，由于初始猜测不充分，可能无法获得解决方案，这可能会限制允许的参数变化的大小。高吉。这项工作的目标是扩展优化算法可以轻松应用的学科领域。为了达到最有效的效果，必须最大限度地减少与重复功能评估相关的计算费用，并最大限度地提高成功导航到所需最佳值的可能性。为了实现这种鲁棒性和效率，我们正在研究结合全局和局部搜索算法、使用函数逼近技术并在分布式计算网络和大规模并行机器上执行的优化策略（如果可行）。
The benefits of applying optimization to computational models are well known, but their range of application to date has been limited. This work attempts to extend the disciplinary areas to which optimization dgorithms may be readily applied through the development and application of advanced optimization strategies capable of handling the computational difficulties associated with complex simulation codes. Towards this goal, a flexible software framework is under continued development for the application of optimization techniques to broad classes of engineering applications, including those with high computational expense and those with nonsmoothness in the design space. Object-oriented software design with C++ has been adopted as a tool to provide a flexible, extensible, and robust multidisciplinary toolkit that establishes the protocol for wrapping parameter optimization around computationally-intensive simulations. The objectoriented approach is well-suited for handling this large software undertaking, in which a myriad assortment of optimization algorithms, approximation techniques, and hybridized strategies must be generically interfaced with broad classes of analysis capabilities. Demonstrations of optimization using the software are presented in fluid mechanics, heat transfer, nonlinear solid mechanics, and combinations thereof. Optimal results are presented along with technical lessons that were learned in the optimization process. *Senior Member of Technical Staff (SMTS), Structural Dynamics Dept., member AIAA. fSMTS, Navigation Guidance Dept., member AIAA. SMTS, Structural Dynamics Dept., member AIAA. §SMTS, Intelligent Systems Dept. W.0. Box 5800, Albuquerque, NM 87185-0439, USA. This work performed at Sandia National Laboratories supported by the U.S. Department of Energy under contract DE-AC04-94AL85000. This paper is declared a work of the U.S. Government and is not subject to copyright protection in the United States. Introduction Computational methods developed in fluid mechanics, structural dynamics, heat transfer, nonlinear large-deformation mechanics, manufacturing and material processes, and many other fields of engineering can be an enormous aid to understanding the complex physical systems they simulate. Often, it is desired to utilize these simulations as virtual prototypes to improve or optimize the design of a particular system. The goal of the effort described in this paper is to enhance the utility of this broad class of computational methods by providing them with a general optimization capability. This fills a critical gap in computational simulation. A significant impact will be to increase the utility of computational methods as design tools for simulating manufacturing processes, tooling, or products. The optimization objectives can be utilized to minimize weight or defects or to maximize performance, reliability, throughput, reconfigurability, agility, or design robustness (insensitivity to off-nominal parameter values). A systematic, rapid method of determining these optimal solutions will result in better designs, decrease hardware expenses, shorten the design cycle, and decrease time to market. Despite the breadth of current computational expertise and capabilities, the impact of computational methods is diminished and ultimately limited by the inability to use these methods to find "optimal" solutions. Towards these ends, we have targeted the needs of a broad class of computational methods in order to provide a general optimization capability. The disciplinary scope of previous optimization work has been limited, and much work has focused on wellbehaved (i.e smooth, convex) problems [1],[2]. The standard optimization techniques used for these wellbehaved problems do not perform well when applied to problems with high computational time and complexity, including those with severely nonlinear, discontinuous, and multimodal parameter spaces. Addressing these Issues extends the range of applications where the benefits of optimization can be realized. TechnicalZssues. Optimization techniques iteratively manipulate a set of parameters to extremize an objective function, subject to a set of constraints. The coupling of optimization with complex computational methods is difficult, and optimization algorithms often fail to converge efficiently, if at all. The difficulties arise from the following traits, shared by many computational methods: 1. The time required to complete a single function evaluation with one parameter set is large. Hence, minimization of the number of function evaluations is vital. 2. Analytic derivatives (with respect to the parameters) of the objective and constraint functions are frequently unavailable. Hence, sensitivity-based optimization methods depend upon numerically generated gradients which require additional function evaluations for each scalar parameter. 3. The parameters may be either continuous or discrete, or a combination of the two. 4. The objective and constraint functions may not be smooth or well-behaved; i.e., the response surfaces can be nonlinear, discontinuous, or even undefined in some regions of the parameter space. The existence of several local extrema (multi-modality) is common. 5. Convergence tolerances in embedded iteration schemes introduce uncertainty (noise) in the function evaluation response surface, which can result in inaccurate numerical gradients. 6. Each function evaluation may require an "initial guess." Function evaluation dependence on the initial guess can cause additional uncertainty in the response surface. Moreover, a solution may not be attainable for an inadequate initial guess, which can restrict the size of the allowable parameter changes. Goaki. The goal of this effort is to extend the disciplinary areas to which optimization algorithms may be readily applied. To be the most effective, one must minimize the computational expense associated with repeated function evaluations and maximize the likelihood of successful navigation to the desired optimum. In order to achieve this robustness and efficiency, we are investigating optimization strategies which combine global and local search algorithms, use function approximation techniques, and execute on distributed computational networks and massively parallel machines (when feasible).