EI、Scopus 收录
中文核心期刊

曲面轨道隔振器参数解耦设计及数据驱动建模

PARAMETER DECOUPLING DESIGN AND DATA-DRIVEN MODELING OF CURVED-SURFACE TRACK VIBRATION ISOLATOR

  • 摘要: 面向工程应用的非线性隔振器研发, 需同时实现动力学特性的精准定制与模型的简约精确重构. 针对第一个问题, 本文提出一种曲面轨道隔振器, 通过逐级设计曲面参数, 可实现隔振器各阶恢复力系数的逐级解耦设计. 针对第二个问题, 提出了频域数据驱动建模方法. 首先将采集到的系统稳态响应展开为傅里叶级数, 进而建立频域信号模型库, 将曲面轨道隔振器的建模问题转化为模型库回归项的稀疏最小二乘问题; 为避免最小二乘“全保留”带来的过拟合, 采用正交向前回归方法(Orthogonal Forward Regression, OFR)筛选模型主导项. 本文通过数值模拟研究了不同外激励频率和幅值的训练数据对模型主导项的影响规律, 结果表明当训练数据包含较宽频带响应时, 所提方法更易得到精确模型. 此外, 傅里叶级数可以很好地分解白噪声, 因而本文所提频域数据驱动方法比时域方法具有更好的抗噪性. 实验进一步验证了方法的可行性, 结果表明本文提出的曲面轨道隔振器参数解耦设计方法和频域数据驱动建模方法具有潜在的工程应用价值.

     

    Abstract: Nonlinear vibration isolators intended for engineering applications must simultaneously achieve precise tailoring of dynamic characteristics and parsimonious yet accurate model construction. Balancing these two requirements remains a critical challenge in the design and analysis of nonlinear isolation systems. To address the first requirement, this paper proposes a curved-track vibration isolator. By hierarchically designing the surface parameters, the restoring-force coefficients of different orders can be independently and progressively decoupled, enabling the flexible and targeted tailoring of nonlinear dynamic characteristics. This design strategy provides a systematic way to match the nonlinear characteristics of the isolator with those of the primary system. To address the second requirement, a frequency-domain data-driven modeling approach is developed. The measured steady-state response is first expanded into a Fourier series, based on which a library of candidate frequency-domain signal models is constructed. The modeling problem of the curved-track isolator is then reformulated as a sparse least-squares regression problem over this model library. To avoid the overfitting issue caused by the full-retention property of conventional Least-Squares Estimation, the OFR (Orthogonal Forward Regression) method is employed to identify the dominant terms of the model, thereby ensuring both model sparsity and accuracy. The numerical simulations are conducted to investigate the influence of training data with different excitation frequencies and amplitudes on the identified dominant terms. The results indicate that training data covering a broader frequency bandwidth facilitate the identification of more accurate models. In addition, since the Fourier series provides an effective decomposition of white noise, the proposed frequency-domain approach exhibits superior noise robustness compared to conventional time-domain methods. Finally, experiments are carried out to validate the feasibility of the proposed methods. The results demonstrate that both the parameter-decoupled design strategy for the curved-track vibration isolator and the frequency-domain data-driven modeling method are effective, and they show strong potential for practical engineering applications.

     

/

返回文章
返回