协方差驱动随机子空间辨识的参数敏感性分析

Parameter Sensitivity Analysis of Covariance-driven Stochastic Subspace Identification

协方差驱动随机子空间辨识(Covariance-driven Stochastic Subspace Identification,SSI-cov)是近年来发展较为成熟的工作模态分析方法。针对其识别精度和效率对参数设置具有较高敏感性的问题,基于敏感性分析方法,利用奇异熵增量跳跃点明显程度、频率平均识别误差、阻尼比总变异系数、振型平均模态置信因子、运行时间5种评价指标以及稳定图,通过一经典五自由度层模型仿真算例,研究Toeplitz矩阵行块数、采样频率、数据长度对SSI-cov识别结果的影响规律。并给出既能满足精度要求又可控制程序运行时间的参数建议取值范围。最后,通过一缩尺三层框架模型在白噪声激励下实测数据对提出的根据SSI-cov改进参数设置进行验证,结果表明提出的各参数建议取值范围均较为合理。

Covariance-driven stochastic subspace identification(SSI-cov)is a relatively mature operational modal analysis method in recent years.However,its identification accuracy and efficiency are highly sensitive to parameter settings.In this paper,the stability diagram and five evaluation indexes,including the apparent degree of the jump point of the singular entropy increment,the average identification error of the frequency,the total variation coefficient of the damping ratio,the average modal confidence factor of the vibration mode and the running time,are employed for the sensitivity analysis.With a classic 5-DOF layer model as the simulation example,the influence of the number of Toeplitz matrix rows,sampling frequency and data length on the recognition result of SSI-cov is studied.The suggestion of value range of parameters that can meet the demands of the precision as well as control running time of the program is given.Finally,the SSI-cov improved parameter setting is verified by the measured data of a scaled three-story frame model.The results show that the proposed parameter range is reasonable.