Abstract: To identify the operational modal parameters of linear slow-time-varying structures,a method based on Moving Window Neighborhood Preserving Embedding(MWNPE)algorithm was proposed.Based on the time freezing theory,the linear time-varying structures were discredited into a finite number of linear time-invariant structures by treating the non-stationary signals in each window as a stationary random sequence with fixed length window.In each window,the neighborhood preserving embedding algorithm was used to find the low-dimensional embedding of the displacement response data in the window,which corresponded to the response matrix of the modal coordinates.The modal natural frequency of the window was identified from the modal response matrix by single degree of freedom recognition technique.The transformation matrix was estimated by Moore-Penrose matrix inverse,which was corresponded to the mode shape matrix.Simulation results of three Degrees of Freedom(DOF)structure showed that the proposed method could effectively identify the operational modal parameters of linear slow-time-varying structures,and the identification accuracy was better than moving window principal component analysis and Moving Window Equivariant Adaptive Separation via Independence(MWEASI).

Key words: linear slow-time-varying structures, operating modal parameters, moving window, neighborhood preserving embedding, moore-penrose matrix inverse

摘要： 为了识别线性慢时变结构的工作模态参数,提出一种基于滑动窗邻域保留投影(MWNPE)的工作模态参数识别方法。该方法基于“时间冻结”理论,利用固定长度的窗口,将每个窗口内的非平稳信号看作平稳的随机序列,从而将线性时变结构离散成有限个线性时不变结构。在每个窗口内,利用邻域保留投影算法寻找窗口内位移响应数据的低维嵌入,低维嵌入与模态坐标响应矩阵相对应;再利用单自由度识别技术从模态响应矩阵中识别出窗口的模态固有频率;最后,利用最小二乘广义逆估计出变换矩阵,变换矩阵与模态振型矩阵相对应。通过质量慢时变三自由度(DOF)的仿真结构验证表明,所提方法能有效识别出线性慢时变结构的工作模态参数,且识别精度优于滑动窗主成分分析方法和滑动窗等变自适应源分离(EASI)方法。

关键词: 线性慢时变结构, 工作模态参数, 滑动窗, 邻域保留投影, 最小二乘广义逆