arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 0806.2748physics.flu-dyn
收藏

激励涡尾流的鲁棒模型识别

Robust model identification of actuated vortex wakes

Jessie Weller, Edoardo Lombardi, Angelo Iollo

摘要:

我们提出了一个低阶的建模技术,驱动流的逆问题的正则化的基础上。逆问题的目的是最小化模型预测和一些参考模拟之间的误差。待识别的参数是多项式展开的系数的子集,该多项式展开对少量全局模式的时间动态进行建模。这些全局模式是通过适当的正交分解找到的,这是一种在能量方面计算现有仿真数据库中最具代表性的元素的方法。结果表明,低阶控制模型的基础上一个简单的Galerkin投影和通常的校准技术是不可行的。它们要么是不适定的,要么是在用于预测不属于原始解数据库的情况下,它们给出了一个差的近似解。相比之下,数值证据表明,我们提出的方法是强大的控制律的变化,从而允许实际使用这种模型进行控制。

英文摘要:

We present a low-order modeling technique for actuated flows based on the regularization of an inverse problem. The inverse problem aims at minimizing the error between the model predictions and some reference simulations. The parameters to be identified are a subset of the coefficients of a polynomial expansion which models the temporal dynamics of a small number of global modes. These global modes are found by Proper Orthogonal Decomposition, which is a method to compute the most representative elements of an existing simulation database in terms of energy. It is shown that low-order control models based on a simple Galerkin projection and usual calibration techniques are not viable. They are either ill-posed or they give a poor approximation of the solution as soon as they are used to predict cases not belonging to the original solution database. In contrast, numerical evidence shows that the method we propose is robust with respect to variations of the control laws applied, thus allowing the actual use of such models for control.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑