Improved operational modal analysis by modified data-driven stochastic subspace identification featuring visualization

Nikka Marie Bucad Sales

doi:10.1016/j.ymssp.2025.113071

Improved operational modal analysis by modified data-driven stochastic subspace identification featuring visualization

2025-07-10

0

PRE

AI

N

Nikka Marie Bucad Sales

C

Chia‐Ming Chang

DOI：10.1016/j.ymssp.2025.113071

Original

Original request for help

Save

Abstract

En 中文

Operational modal analysis is beneficial to structural health monitoring and design. Data-driven stochastic subspace identification, in particular, is one of the commonly accepted system identification methods in operational modal analysis due to stability, accuracy, and efficiency. This parametric method requires appropriate estimation of the model order for extracting accurate modal parameters. A stabilization diagram is an alternative approach to address this challenge. Still, establishing this tool is computationally expensive and highly dependent on the experience of users. An improved modal parameter estimation framework is therefore developed in this study by modifying the data-driven stochastic subspace identification to effectively identify accurate structural modes, while not compromising the ease of usage. In this framework, due to the application of singular value decomposition to the projection matrix, the observability matrix is sorted into interpretable components. These components are paired based on vector direction similarity on the basis of modal energy, wherein resulting pairs are separately utilized for modal parameter estimation, allowing for the sequential identification of modes. Additionally, a mode validation algorithm is developed to eliminate spurious modes. A numerical study is carried out to demonstrate the proposed framework, and two experimental examples are conducted to investigate the performance. The row size of the Hankel matrix, as a user-defined parameter, is found to influence the precision of estimates and mode identifiability, where non-identification can be mitigated through adjustment without overestimation. Results further verify the reliance of the proposed framework on the modal energy of structural modes, its accuracy for lightly damped systems, and its reduced computational cost through its operations on smaller matrices at a single model order only. Moreover, numerical and experimental studies illustrate its ease of usage through the sequential mode selection process, which required minimal expert intervention, and through verifiable outcomes via visualization of the modal responses.

Improved operational modal analysis by modified data-driven stochastic subspace identification featuring visualization

Abstract

Journal

Organization

Cited Papers