A machine learning-based stochastic subspace approach for operational modal analysis of civil structures

Abstract

Based on two machine learning algorithms and eight evaluation indexes, this paper proposes a novel stochastic subspace identification (SSI) approach for the operational modal analysis of civil structures. The detailed procedure of the proposed approach is first demonstrated via a numerical example of a five-degree-of-freedom (5DOF) simulation model, and the results show that the proposed approach can automatedly and effectively distinguish the physical modes from non-physical ones for the SSI. The modal parameters of the 5DOF model estimated by the proposed approach agree well with their theoretical values, verifying the accuracy and effectiveness of the proposed approach. Furthermore, the proposed approach is applied to field measurements on a 195-m-tall building under its operating condition, and the modal parameters of the building are well identified in an automated manner. Through the statistical analysis, the probability distributions and amplitude-dependent features of the natural frequencies and damping ratios of the 195-m-tall building are further revealed. The objective of this study is to provide an efficient tool for the operational modal analysis of civil structures.