A Comparative Study of the Data-Driven Stochastic Subspace Methods for Health Monitoring of Structures: A Bridge Case Study

Abstract

Subspace system identification is a class of methods to estimate state-space model based on low rank characteristic of a system. State-space-based subspace system identification is the dominant subspace method for system identification in health monitoring of the civil structures. The weight matrices of canonical variate analysis (CVA), principle component (PC), and unweighted principle component (UPC), are used in stochastic subspace identification (SSI) to reduce the complexity and optimize the prediction in identification process. However, researches on evaluation and comparison of weight matrices’ performance are very limited. This study provides a detailed analysis on the effect of different weight matrices on robustness, accuracy, and computation efficiency. Two case studies including a lumped mass system and the response dataset of the Alamosa Canyon Bridge are used in this study. The results demonstrated that UPC algorithm had better performance compared to two other algorithms. It can be concluded that though dimensionality reduction in PC and CVA lingered the computation time, it has yielded an improved modal identification in PC.

Keywords

• structural health monitoring (SHM)
• subspace system identification (SSI)
• principle components (PC)
• unweighted principle components (UPC)
• canonical variate analysis (CVA)