Application of covariance statistical method for damage identification on railway truss bridge using acceleration response: experimental and numerical validation

This paper presents a novel statistical analysis-based approach to non-parametric damage detection in truss bridges. The method utilizes the normalized acceleration response time histories (NARTHs) of a bridge under random excitation. The coefficients of variation matrices are calculated using NARTHs for the truss bridge in both its baseline and damaged states. The results are shown as the difference in covariance matrices (DCMs) between the two conditions (pristine and damaged). The row or column-wise summation of the DCM matrix yields a summation of the difference in covariance matrix (SDCM) bar plot having one distinct peak for the damaged member. The location and relative severity of the damage can be determined by examining the DCM matrix and the SDCM bar plot. The variation in magnitude of the coefficients demonstrates the relative severity of damage. Experimental investigation of the proposed method for detecting damage on the truss bridge shows promising results. The approach is then numerically validated using a finite element model of a truss bridge. The proposed method effectively identified, located, and evaluated damage, even when the acceleration time histories are contaminated with noise. The case of multiple-member damage detection in the truss structure is further investigated in the study. The early detection of damage and monitoring of its progression through this method can assist in creating efficient maintenance strategies for truss bridge infrastructure.