An Integrated novel random decrement technique and system identification framework for modal features extraction in structural health monitoring

Structural health monitoring (SHM) techniques are used for monitoring the condition of structures, including bridges, buildings, dams, among others, through data acquired from embedded sensors. An integral aspect of SHM involves feature extraction, wherein algorithms analyze raw sensor data to generate explainable features regarding the health state of structures. The random decrement technique (RDT) is a notable signal processing method used to estimate the free response of mechanical structures, crucial for extracting frequency features related to structural dynamics. However, studies have shown that RDT encounters performance issues in cases of nonlinear damping, significant signal noise, and overlapping vibration modes. In a previous study, a data-driven classifier for transient and stationary regimes in multivariate dynamic systems was introduced, employing differential geometry tools to analyze stochastic properties. In this paper, the authors applied this classifier to develop a novel frequency-feature extraction method for SHM. This method incorporates a modified triggering condition for RDT and a recent system identification technique, aiming to improve frequency-feature estimation, especially in segment selection and modal parameter estimation of free responses in non-linear structures. Validation of the proposed approach employed mechanical vibration data from four accelerometer sensors on the “Andoian DF 12” bridge in Guipuzkoa, Spain, covering a one-year period and providing comprehensive insights into structural dynamics across different seasons. Comparative analysis against traditional RDT and various system identification techniques demonstrated that the proposed methodology improve the estimation of structural free responses by reducing noise and requiring fewer data compared to traditional RDTs. Moreover, the estimated modes aligned with the spectral content of mechanical sensors, indicating suitability for online applications with minimal computational costs and accurate modal parameter estimation for SHM in structures.